Patent Grant
10542961
The present application incorporates the following material by reference, in its entirety: Ohad BarSimanTov, Ph.D. Dissertation, Binghamton University (2014, embargoed).
Cardiac Output (CO) is the one of the most important measurements for assessing human health. CO refers to the volume (in liters) of blood pumped out by the left ventricle to the body per minute. Unfortunately, the lack of accuracy and difficultly of measuring this parameter in the primary care and home care environments has drastically limited its use. As a result, other more readily obtained assessments are commonly used to provide an estimation of CO, for example, heart rate, respiratory rate, blood pressure, body temperature (i.e., the vital signs).
A vital signs assessment of an individual's health is undertaken during any hospital admission, any doctor visit, by a medic, or at the start of almost any healthcare intervention. The reason for this assessment is to check how well the vital organs of the body, lungs, heart, etc., are working, and thereby identify any early signs of illness. There are four Vital Signs commonly used to quickly assess the health status of an individual [1]; pulse rate, blood pressure, breathing rate, and body temperature. The pulse rate also is called heart rate, which is the number of heart beats per minute. Heart Rate (HR) can readily be measured by sensing the pulsations in an artery. Blood pressure is typically measured at the arteries of the arm and represents the driving force of blood through the arterial system. The systolic pressure is the high pressure value, and indicates the heart's contraction force. The diastolic pressure is lower pressure value and happens during the relaxation phase of the heart. The blood pressure and pulse rate provides only a very crude estimate of Cardiac Output (CO). Cardiac Output is commonly compared to the person's body surface areas (BSA) which can be obtained through the person's height and weight. Cardiac Index (CI) is the ratio of CO to BSA and is relatively constant across the population in healthy individuals, though it decays with increasing age. The third vital sign is breathing rate and is obtained by counting the number of times the person's chest rises and falls in one minute. Usually, a physician will pay attention to abnormal sounds during respiration, which can provide information about the person's respiratory condition. The last vital sign is the body temperature, which provides a measure of the individuals' thermo-regulatory status, a critical aspect of health in a homoeothermic organism.
Introduction of a dye into the blood stream provides one method to measure CO. The Fick method uses two catheters to extract blood samples from the pulmonary artery and from the aorta and measure the oxygen levels in both. The dye dilution uses a single catheter to extract blood samples from the aorta after the dye was injected in the median antecubital vein. Around 1960s, the dye dilution came to be viewed as the most accurate technique to measure CO. In 1970s the pulmonary artery catheter (PAC) was introduced and became the ‘gold standard’ apparatus as it could be used in both dye-dilution and thermo-dilution methods, which involves just one catheter to inject a solution to the right atrium and measure its change in the pulmonary artery. Since then, invasive techniques using PAC, and other arterial catheters, are widely used in hospital settings. However, these invasive methods and techniques introduce significant risk of deleterious complications such as bleeding, pneumothorax, and infection and widespread use of these devices is falling out of favor. Therefore, non-invasive CO monitoring measurements are preferred by physicians and patients.
Numerous non-invasive cardiac monitoring technology have been developed over the years to address the need for non-invasive cardiac monitoring, including Electrocardiogram (ECG), Computed Tomography (CT), Magnetic Resonance Imaging, Doppler Ultrasound, Echocardiogram, and Bio-impedance. All of these improve patient safety and significantly reduce complications. Some of them also dramatically reduce the cost of cardiac monitoring. ECG does not provide any specific information about the mechanical activity (i.e. pumping activity) of the heart, only the electrical activity. Similarly, CT and MRI provide only static images of the heart. Doppler Ultrasound and Bioimpedance provide estimates of the volume of blood ejected by the heart during each contraction, which when multiplied by heart rate, provides an estimate of CO.
The total volume of blood that the heart pumps during each contraction is referred to as the stroke volume. Cardiac output (CO) is simply the product of the heart rate and stroke volume. Stroke volume depends on atrial filling pressure, the end diastolic pressure in the ventricle (backpressure), and the contractility of the ventricle. Typically, a healthy cardiac muscle will eject between 55% and 70% of the ventricular volume, or approximately 70 ml [23]. As the cardiovascular system is a closed system, the right ventricle volume and ejection fraction roughly equal to those on the left ventricle at any given point in time. A complete cardiac cycle occurs during each heartbeat, both atria simultaneously contract followed by contraction of the ventricles [24]. During contraction, the blood pressure in the left ventricle varies rapidly from about 80 mmHg (typical diastolic pressure) to about 120 mmHg (typical systolic pressure). The right ventricle pressures are much lower, ranging from 2 mmHg to 30 mmHg. As all tissues are critically dependent on adequate blood perfusion to maintain normal function, a measure of cardiac output can provide important information regarding the cause of any tissue or organ system failure. Based on an individual's gender, age, and training levels, HR and SV differ, but the heart supplies the needed amount of blood to the entire body by regulating these two factors. If one component is lower the other adapts to a higher level during either rest or exercise. Because HR can vary over a range of 2-3 fold, and SV can vary over a range of 2-3 fold, the heart has the capability to increase its output by up to roughly tenfold. The ability to attain this full operating range is the essence of athletic training, therefore, accurate, non-invasive, continuous CO monitoring has major applications not just in the healthcare domain, but also in the athletic domain.
The heart generates sounds during each cardiac cycle due to the flow of blood through the heart; four heart sounds are commonly defined. The first two, S1 and S2, are normal and associated with the closing of the mitral and tricuspid valves and the aortic and pulmonary valves during the beginning of systolic and diastolic phases respectively [37]. These heart sounds can provide a basis for synchronization of heart contraction timing. The third and the fourth heart sounds, S3 and S4, may be normal or abnormal and are the result of turbulent flow through the values, or regurgitation back through valve. These latter heart sounds are often referred to as murmurs.
Following the ECG P wave, contraction of the atria is associated with left ventricular volume increase and small left ventricle and left atrium pressure increase. Following The ECG QRS complex the ventricles contract and the left ventricle pressure increases rapidly. This causes the closing of the tricuspid valve, which produces the first heart sound, representing the beginning of the systolic phase. The left ventricle pressure continues to increase and the aorta semilunar valve opens when the left ventricle pressure exceeds the aorta pressure. As a result, the aortic pressure is increased and the left ventricle volume decreases. The blood from the left ventricle is expelled into the aorta until the relaxation phase begins. At this stage, the aortic valve closes and produces the second heart sound which is the end of the systole phase and the beginning for the diastole phase. The left ventricle continues to relax, and then the tricuspid valve opens and oxygenated blood enters to the left ventricle. These flows and volumetric changes produce forces, and motions of the chest wall, which have physiological correlates.
In the 19 century it was first proposed that measurements of chest wall movement during cardiac muscle contraction could serve as a useful measure of stroke volume. Many techniques and methods have been developed to capture cardiac activity non-invasively using sensing techniques, including Sphygmography, Apex-cardiogram, Mechanocardiogram, Vibrocardiogram, accelerocardiogram, Cardiokymogram, Ballistocardiogram, Kinetocardiogram, Seismocardiogram, and other related methods were introduced to measure cardiac activity from body and/or chest motion. All of these methods measure a mechanical motion below 20 Hz, which is at or below the audible range of human hearing. Acoustic signals below the range of human hearing are referred to as infrasonic [64].
The Vibrocardiogram (VbCG) provides for measurement of the phases of isometric contraction, ejection, systole and diastole obtained from the precordial vibration tracing, with an accuracy as good as that realized with direct cardiac catheterization [79, 80]. Those recordings permit the total energy spectrum of the heart in a displacement form, like the ACG. However, the VbCG signal, is obtained using a microphone, which provides a measure of the velocity of the chest motion. If this signal is integrated, it provides the displacement of the chest wall. The VbCG waveform typically consists of four main deflections [79]. The segment between J2 and L is considered to be the ejection time (ET) and the isometric contraction time (ICT) is considered to be between R left and J2. However, in later studies it has been found that the isometric contraction is happening after the ECG R way. The comparison between the dye dilution and VbCG has been explored and a good correlation of the ejection time has been found (r=0.9). Also, an estimation of cardiac output has been shown to be possible [81].
The recording location of the VbCG is defined as the area between the 4th to the 5th left intercostal spaces near the sternum, similarly to ACG, accelerocardiogram [82], and Cardiokymography (CKG) [84]. The investigation of heart diseases using these methods has also been described [81, 83]. VbCG and Accelerocardiogram record the acceleration of the chest wall at the apex location (behind the fifth left intercostal space, 8 to 9 cm from the mid-sternal line, slightly medial to the midclavicular line.) ACG and CKG record the displacement of the chest wall at the apex location. These four methods require breath holding while the subject is lying down. Moreover, the signal varies due to the position of the transducer on the chest, though traditionally the sensor is located exclusively at the apex, and the interpretation of the signal at other locations is not generally interpreted. Some of the recent publications have used optical sensing devices to measure cardiac activities [85, 86] and some investigators have concluded that these techniques have the capability and certain advantages for long term cardiac monitoring [87].
An alternative method of recording precordial (the region of the chest over the heart) movements is the Kinetocardiogram (KCG) [103, 104, 105]. Like the ACG and BCG, KCG records the displacement of a human body as a result of cardiac activity. This method was very similar to ACG, but the recordings were taken at specific known locations on the chest like the V1-V6 in ECG, and included the apex location. The recordings utilize a sensitive piezoelectric transducer (accelerometer) during the end of normal expiration [103]. The recordings demonstrate a range of motion over the chest of about two hundred microns.
Seismocardiogram (SCG) is a similar method which records the acceleration of the chest wall; however, the recording location is at the sternum, similarly to one of the KCG recording locations. SCG was first described in 1961 [107]. In 1990 SCG recordings were performed using an accelerometer and microprocessor based signal analysis device [108]. The frequency range of the SCG signal was found to be 0.3 to 50 Hertz [108]. Moreover, this method has been investigated and compared to left ventricular function [109] and the relationship between SCG and Echocardiogram [110].
The idea of integrating a portable Seismocardiography signal to measure CO has therefore existed for quite some time [115, 116, 117]. However, it is not fully developed.
There are many mathematical tools used to analyze a signal in the frequency domain. The traditional Fourier and Laplace methods work well in smooth continuous signals, for which the statistics properties do not vary in time (i.e. the signals are stationary). They are widely used in science and engineering and suitable when the recorded signal is stationary. The basis functions, however, are sine and cosine functions, which extend over infinite time intervals and are therefore not suited for non-stationary signals [122]. Short Time Fourier Transform (STFT), which is used to analyze small segments of a signal, is more suitable for non-stationary signals if the statistical properties of the signal vary slowly relative to the dominant frequency of the signal. In the context of infrasonic vibrations associated with heart activity, the received signal consists of many frequencies which are very close to frequency of variation (˜1 Hz, i.e. the heart rate frequency) and is not suitable for the use of STFT. Wavelet analysis is a tool which can be used to extract spectral information where ST Fourier methods fail [124]. Wavelet analysis transforms signals into a two dimensional time-frequency domain, utilizing a series of convolution operations on the signal against particular filters at various positions and time scales. This process not only separates high frequency component from low frequency components, but as well, the wavelet scaling process allows one to look at a signal within a small window, and notice the time location of small and high frequency features. In addition, noises can be subtracted fairly easily from the recorded signal, which makes this tool very powerful in many applications: data compression, signal and image processing, music analysis, speech discrimination, quantum physics, optics, earthquake-prediction, radar, human vision, seismic geology, and more. The mathematical proof of the transformations is provided in the papers [124-126].
Wavelet analysis typically consists of three major stages. The first is decomposition, which separates the signal's high frequency components from the low frequency components. This process continues recursively through different stages and provides signal analysis while convolving the signal with high pass and low pass filters. Based on the “cost” of each decomposition level the efficiency of the analysis is calculated to ensure the desired information is extracted from the signal. The second stage is called threshold, where each of the decomposed segments, called packets, is processed via mathematical computations. The last stage is signal reconstruction, which collects the packets and reconstructs the desired signal. This three step process is very common in data compression, and image processing.
The set of high pass and low pass filters are scaled functions of the Mother Wavelet. A Mother Wavelet is a wavelet prototype function which is adapted (scaled and shifted) during the procedure of wavelet analysis. The analysis is performed with a high pass filter and a low pass filter using the Mother Wavelet coefficients. Different Mother wavelets will produce different outputs and different signal information [126]. Therefore, a more suitable mother wavelet function will provide better information extraction and better noise reduction from the original signal.
The original signal can be reconstructed using all scalings and details of the decomposed signal with knowledge of the decomposed mother wavelet. Wavelet decomposition separates the signal to details g[n] and scaling h[n] coefficients with the scaling and details decimated by factors of two. Each level of decomposition can be obtained from the scaling factor and separated into scaling and details again. Similarly, the reconstruction process can use the scaling and detail packets to reconstruct the previous decomposition level packets until the original is reconstructed. The scaling h[n] and detail g[n] are convolved with the Mother wavelet until the original signal is reconstructed.
Wavelet analysis has become one of the most commonly used digital signal processing tools, with applications in data compression, image processing, time series data filtering, material detection and de-noising [125, 189]. Over the past two decades, the use of wavelet analysis has increased rapidly in the biomedical field with analyses being applied to remove base line variation and high frequency components from the electrocardiogram (ECG) and to distinguish specific features within the ECG waveform [153, 154]. Wavelets have also been used in analysis of electromyograms (EMG) [155, 156], mechanomyograms (MMG) [157], electroencephalograms (EEG) [158, 159], seismocardiograms [115], and other medical applications. Wavelet analysis is widely used to de-noise data and to separate observed components where decomposition, thresh-holding, and reconstruction are computed. Of course, physiological recordings are not signals (i.e. messages) per se. In communication systems, the original transmitted message is known and can be compared to the received signal; physiologic recordings can only be interpreted based on a set of assumptions regarding the performance of the physiologic system, rather than comparing to a-priori signal. Accordingly, the appropriate processing algorithm must be identified by correlating the output produced by various analyses to some system characteristic of interest.
See, e.g., U.S. Patent and U.S. Pat. Nos. 9,008,367, 9,008,754, 9,008,762, 9,011,338, 9,014,453, 9,014,789, 9,014,790, 9,014,809, 9,020,585, 9,026,193, 9,026,200, 9,028,405, 9,028,407, 9,031,301, 9,031,642, 9,031,649, 9,033,883, 9,033,887, 9,033,893, 9,037,223, 9,037,224, 9,042,619, 9,042,952, 9,042,958, 9,044,144, 9,044,149, 9,044,171, 9,044,558, 9,049,981, 9,049,985, 9,050,007, 9,050,014, 9,055,871, 9,055,884, 9,056,172, 9,060,669, 9,060,683, 9,060,689, 9,060,695, 9,060,722, 9,060,733, 9,066,672, 9,066,679, 9,066,680, 9,066,686, 9,066,708, 9,069,130, 9,072,437, 9,072,438, 9,075,446, 9,076,202, 9,079,060, 9,081,148, 9,084,531, 9,084,576, 9,084,611, 9,086,467, 9,087,368, 9,089,269, 9,092,691, 9,095,266, 9,095,313, 9,095,505, 9,097,756, 9,101,286, 9,101,772, 9,107,571, 9,107,584, 9,107,586, 9,107,623, 9,107,624, 9,113,794, 9,113,795, 9,113,826, 9,113,830, 9,114,260, 9,116,835, 9,119,547, 9,119,573, 9,125,574, 9,125,577, 9,125,578, 9,126,050, 9,131,852, 9,131,864, 9,131,902, 9,136,980, 9,138,150, 9,144,394, 9,147,268, 9,149,219, 9,149,231, 9,149,244, 9,149,645, 9,155,482, 9,155,484, 9,155,893, 9,161,705, 9,162,074, 9,163,216, 9,167,971, 9,168,380, 9,168,385, 9,168,419, 9,171,353, 9,173,566, 9,173,579, 9,173,909, 9,174,061, 9,175,095, 9,176,319, 9,178,330, 9,179,890, 9,180,043, 9,180,300, 9,183,351, 9,183,626, 9,186,066, 9,186,067, 9,186,068, 9,186,079, 9,186,105, 9,186,106, 9,186,521, 9,192,328, 9,192,336, 9,192,446, 9,197,173, 9,198,582, 9,198,586, 9,198,604, 9,198,616, 9,198,634, 9,199,078, 9,201,902, 9,202,008, 9,204,796, 9,208,173, 9,208,557, 9,208,587, 9,209,782, 9,211,413, 9,215,298, 9,215,980, 9,215,987, 9,215,991, 9,216,001, 9,216,065, 9,220,440, 9,220,455, 9,220,459, 9,220,460, 9,220,467, 9,220,856, 9,226,660, 9,226,665, 9,226,676, 9,236,046, 9,237,855, 9,239,951, 9,241,667, 9,242,090, 9,245,091, 9,247,901, 9,248,288, 9,248,306, 9,249,200, 9,254,089, 9,254,093, 9,254,095, 9,254,102, 9,254,383, 9,258,561, 9,259,167, 9,259,591, 9,261,573, 9,262,826, 9,264,877, 9,267,936, 9,269,127, 9,277,871, 9,277,956, 9,278,226, 9,282,896, 9,282,902, 9,282,908, 9,282,910, 9,282,911, 9,282,925, 9,282,931, 9,286,662, 9,289,133, 9,289,136, 9,289,150, 9,289,165, 9,289,167, 9,289,545, 9,294,074, 9,295,391, 9,301,703, 9,301,705, 9,302,111, 9,304,121, 9,305,334, 9,305,350, 9,307,407, 9,307,917, 9,308,042, 9,308,052, 9,314,181, 9,314,305, 9,319,028, 9,319,212, 9,320,491, 9,324,005, 9,324,141, 9,324,144, 9,326,682, 9,326,697, 9,326,722, 9,327,130, 9,330,092, 9,330,459, 9,332,357, 9,332,939, 9,332,942, 9,339,202, 9,339,206, 9,339,230, 9,339,241, 9,339,436, 9,339,662, 9,340,589, 9,341,783, 9,345,413, 9,345,609, 9,345,888, 9,351,640, 9,351,642, 9,351,649, 9,351,661, 9,351,668, 9,351,674, 9,352,057, 9,352,165, 9,354,115, 9,356,731, RE45512, RE45725, RE45922, U.S. Pat. Nos. 9,026,202, 9,026,199, 9,011,346, 8,998,830, 8,992,434, 8,983,854, 8,979,731, 8,979,730, 8,942,779, 8,905,928, 8,882,684, 8,880,156, 8,870,780, 8,862,231, 8,858,449, 8,840,564, 8,834,364, 8,827,918, 8,821,418, 8,798,726, 8,798,714, 8,790,264, 8,764,653, 8,747,312, 8,734,360, 8,731,646, 8,706,204, 8,700,137, 8,696,569, 8,679,034, 8,679,030, 8,603,010, 8,594,787, 8,585,607, 8,562,526, 8,540,651, 8,517,953, 8,509,881, 8,491,492, 8,482,416, 8,475,367, 8,403,865, 8,376,964, 8,376,954, 8,262,582, 8,249,698, 8,024,044, 7,785,257, 7,715,894, 7,660,632, 7,417,536, 7,396,331, 7,314,451, 7,286,871, 7,269,537, 7,077,810, 7,043,293, 7,019,641, 6,993,378, 6,840,907, 6,547,743, 6,535,754, 6,478,744, 6,370,481, 6,179,783, 6,152,879, 6,053,872, 6,050,950, 5,964,720, RE35122, 5,445,162, 5,197,490, 4,934,372, 4,928,692, 4,926,866, 4,911,167, 4,895,155, 4,893,633, 4,889,130, 4,889,123, 4,884,578, 4,848,350, 4,838,275, 4,836,215, 4,817,610, 4,802,486, 4,738,264, 4,681,098, 4,679,569, 4,657,025, 4,519,395, 4,195,643, 4,036,215, 3,960,140, 9,028,407, 9,028,405, 9,026,214, 9,026,201, 9,014,809, 9,002,453, 9,002,446, 8,998,820, 8,996,110, 8,996,107, 8,968,195, 8,961,185, 8,954,146, 8,942,799, 8,934,970, 8,923,965, 8,918,172, 8,909,329, 8,888,710, 8,886,311, 8,862,226, 8,818,524, 8,798,731, 8,792,998, 8,768,461, 8,764,651, 8,750,992, 8,750,981, 8,747,336, 8,747,313, 8,738,119, 8,738,111, 8,727,978, 8,708,903, 8,702,616, 8,700,146, 8,684,925, 8,684,922, 8,684,900, 8,680,991, 8,652,038, 8,649,853, 8,634,930, 8,630,707, 8,626,281, 8,620,427, 8,620,426, 8,617,082, 8,606,349, 8,594,805, 8,588,906, 8,586,932, 8,583,233, 8,583,230, 8,583,229, 8,532,770, 8,531,291, 8,525,687, 8,525,673, 8,515,535, 8,515,534, 8,509,890, 8,500,636, 8,494,829, 8,475,368, 8,473,068, 8,473,055, 8,461,988, 8,456,309, 8,452,389, 8,449,471, 8,445,851, 8,425,415, 8,423,125, 8,401,640, 8,380,308, 8,346,360, 8,328,718, 8,326,429, 8,326,428, 8,323,189, 8,323,188, 8,321,003, 8,301,241, 8,287,459, 8,285,373, 8,275,463, 8,255,042, 8,251,911, 8,244,355, 8,223,023, 8,214,033, 8,209,010, 8,160,701, 8,145,304, 8,121,673, 8,115,640, 8,108,036, 8,103,333, 8,073,541, 8,027,724, 8,025,624, 8,010,194, 8,005,543, 8,000,773, 7,963,925, 7,953,479, 7,949,399, 7,949,398, 7,908,004, 7,877,146, 7,869,876, 7,869,869, 7,856,268, 7,850,616, 7,848,816, 7,846,104, 7,813,805, 7,797,050, 7,787,946, 7,774,055, 7,769,446, 7,733,224, 7,689,283, 7,676,266, 7,672,728, 7,670,295, 7,650,189, 7,636,600, 7,577,478, 7,574,255, 7,558,622, 7,546,161, 7,539,533, 7,539,532, 7,494,459, 7,460,909, 7,338,436, 7,248,923, 7,206,636, 7,177,686, 7,139,609, 7,079,896, 7,039,462, 6,647,289, 6,551,252, 6,477,406, 6,370,424, 6,171,263, 6,155,976, 6,022,322, 5,782,884, 5,718,720, 5,507,785, 20160151013, 20160148531, 20160143594, 20160143543, 20160140834, 20160135754, 20160135705, 20160128641, 20160128630, 20160120464, 20160120434, 20160120433, 20160120431, 20160120418, 20160114169, 20160114168, 20160114162, 20160113618, 20160106378, 20160100817, 20160100803, 20160100787, 20160098835, 20160095531, 20160087603, 20160081573, 20160081561, 20160074667, 20160073965, 20160073959, 20160067433, 20160066881, 20160066860, 20160066799, 20160066788, 20160058308, 20160058301, 20160051822, 20160051233, 20160051205, 20160051203, 20160045123, 20160038091, 20160038038, 20160038037, 20160034634, 20160033319, 20160023013, 20160022999, 20160022166, 20160022164, 20160022156, 20160008613, 20160007932, 20160007907, 20160001089, 20160000350, 20160000346, 20150374983, 20150374300, 20150366532, 20150366518, 20150366511, 20150362360, 20150359492, 20150359486, 20150359467, 20150359452, 20150359441, 20150352369, 20150352367, 20150351699, 20150342490, 20150342488, 20150342478, 20150305632, 20150297907, 20150297112, 20150297111, 20150297104, 20150290453, 20150289807, 20150289777, 20150286779, 20150283027, 20150272464, 20150272457, 20150269825, 20150265348, 20150265175, 20150265174, 20150265164, 20150265161, 20150257715, 20150257712, 20150257700, 20150257671, 20150257668, 20150251012, 20150246195, 20150245782, 20150238148, 20150238147, 20150238106, 20150238091, 20150223863, 20150223756, 20150223733, 20150223711, 20150223708, 20150220486, 20150216483, 20150216480, 20150216433, 20150216431, 20150216426, 20150208939, 20150204559, 20150201859, 20150201854, 20150201853, 20150196256, 20150196213, 20150190637, 20150190636, 20150190088, 20150190060, 20150182160, 20150178631, 20150174307, 20150173631, 20150171998, 20150165223, 20150164433, 20150164375, 20150164358, 20150164355, 20150164349, 20150164340, 20150164339, 20150157387, 20150157269, 20150157258, 20150157239, 20150157218, 20150148635, 20150142070, 20150142069, 20150141863, 20150141861, 20150141860, 20150141857, 20150141846, 20150137988, 20150135310, 20150134019, 20150133807, 20150133795, 20150127067, 20150127061, 20150126876, 20150126848, 20150126833, 20150125832, 20150122018, 20150119711, 20150117741, 20150112606, 20150112452, 20150112409, 20150112220, 20150112212, 20150112211, 20150112209, 20150112208, 20150112159, 20150112158, 20150112157, 20150112156, 20150112155, 20150112154, 20150109124, 20150106020, 20150105695, 20150105681, 20150105631, 20150103360, 20150099946, 20150099941, 20150088214, 20150088016, 20150088004, 20150087947, 20150087931, 20150086131, 20150080746, 20150073234, 20150068069, 20150065835, 20150065815, 20150065814, 20150057512, 20150051452, 20150046095, 20150045684, 20150038856, 20150032178, 20150031969, 20150031964, 20150025394, 20150025393, 20150025336, 20150025335, 20150025334, 20150025328, 20150018637, 20150018632, 20150016702, 20150005655, 20150005594, 20150005646, 20140378849, 20140371574, 20140362013, 20140350361, 20140323821, 20140309543, 20140308930, 20140288551, 20140288442, 20140277241, 20140277239, 20140275925, 20140275886, 20140275829, 20140275824, 20140267299, 20140266787, 20140249429, 20140222115, 20140221859, 20140221786, 20140207204, 20140194702, 20140163425, 20140163368, 20140163343, 20140151563, 20140143064, 20140142451, 20140142444, 20140142437, 20140135645, 20140135634, 20140128953, 20140121476, 20140114370, 20140104059, 20140094875, 20140088676, 20140077946, 20140066798, 20140055284, 20140052209, 20140046209, 20140039333, 20140039330, 20140012144, 20140012099, 20140005496, 20130345591, 20130338460, 20130331904, 20130331661, 20130310700, 20130296962, 20130261473, 20130245722, 20130245502, 20130231574, 20130211482, 20130211291, 20130211271, 20130204122, 20130197597, 20130197375, 20130197322, 20130190835, 20130190645, 20130178718, 20130172691, 20130171599, 20130165819, 20130158415, 20130144178, 20130137998, 20130123873, 20130109989, 20130095459, 20130085401, 20130072807, 20130069780, 20130060297, 20130060296, 20130053926, 20130053913, 20130053912, 20130053907, 20130030487, 20130030486, 20130030484, 20130030315, 20130030314, 20130030312, 20130023957, 20130023956, 20130009783, 20130006317, 20120330373, 20120330109, 20120296228, 20120277545, 20120271382, 20120271371, 20120271177, 20120253419, 20120245476, 20120245464, 20120242501, 20120238800, 20120226126, 20120220835, 20120203090, 20120203077, 20120197333, 20120185012, 20120179216, 20120172742, 20120172741, 20120165892, 20120157861, 20120157856, 20120157798, 20120143072, 20120136263, 20120132211, 20120123279, 20120095357, 20120095352, 20120092157, 20120092156, 20120071792, 20120022844, 20120022384, 20120022350, 20120022336, 20120010677, 20110319782, 20110319778, 20110319776, 20110301660, 20110263994, 20110251502, 20110208016, 20110196442, 20110196441, 20110196254, 20110181422, 20110172500, 20110160656, 20110152974, 20110137110, 20110130671, 20110130670, 20110118614, 20110115624, 20110112442, 20110106558, 20110105930, 20110098770, 20110098583, 20110087115, 20110087113, 20110082511, 20110066205, 20110066042, 20110066041, 20110060235, 20110060230, 20110046508, 20110046498, 20110036801, 20110034811, 20110021928, 20110015704, 20110015703, 20110015702, 20110015468, 20100331908, 20100305634, 20100304864, 20100274219, 20100256701, 20100249628, 20100210921, 20100204550, 20100179438, 20100169810, 20100169122, 20100152795, 20100123587, 20100121406, 20100114207, 20100113944, 20100100150, 20100100148, 20100094147, 20100094102, 20100069768, 20100030090, 20100010556, 20090318987, 20090318779, 20090312612, 20090270746, 20090227877, 20090227876, 20090203972, 20090105556, 20090099473, 20090078875, 20090076401, 20090076350, 20090076349, 20090076348, 20090076346, 20090076343, 20090076342, 20090054758, 20090054742, 20090036940, 20090024044, 20080312523, 20080306397, 20080275349, 20080275314, 20080269625, 20080262367, 20080230705, 20080230702, 20080194975, 20080161877, 20080128626, 20080119749, 20080091114, 20080042067, 20080039904, 20080013747, 20080004904, 20080004672, 20080001735, 20070299349, 20070276270, 20070273504, 20070265533, 20070260285, 20070191742, 20070150006, 20070118054, 20070106170, 20070103328, 20070083243, 20070083128, 20070013509, 20060293607, 20060247542, 20060241510, 20060211909, 20060206159, 20060167334, 20060149139, 20060142634, 20060111754, 20060094967, 20050234289, 20050137480, 20050124864, 20050113666, 20040267086, 20040097802, 20030233132, 20030233034, 20030135127, 20030135097, 20030045806, 20020040192, 20020032386, each of which is expressly incorporated herein by reference in its entirety. These references disclose, for example, complementary technologies, aspects and details of implementation, applications, and the like.
The Mother Wavelet function is often selected based on the shape and characteristic of the feature one is trying to extract. Some functions are better at capturing amplitude and phase changes; others are better at synthesizing data and quantitative information. Domingues et al. [190] and Chourasia et al. [191] show examples, where selection of a particular mother wavelet provides better feature extraction than others. Rather than accepting such a trade-off by selecting a single basis set, it should be possible to combine information from multiple mother wavelets. See, U.S. Pat. No. 8,874,477, expressly incorporated herein by reference.
If one has inadequate a-priori understanding of the characteristics which need to be extracted for a particular application there may be advantages in performing multiple full tree decompositions using multiple mother wavelets, and then recombining specific packets to create a hybrid. While encouraging in principle, this approach soon faces the curse of dimensionality; the number of combinations increases factorially. Genetic Algorithms (GAs) have some ability to deal with combinational exploration, so this approach was explored.
Genetic Algorithms (GAs) and Wavelets have been combined recently in image processing for fault detection [160, 161], voice recognition [162], and other applications [163, 164], but these investigations used a binary encoding for packet selection. Previous investigations [149] imply that a better approach is to incorporate an index representation (genes are the indexes of the features to select from a possibly large pool of features), with a special subset selection crossover operator. This approach has been used in medical imaging, and also genomic and proteomic data mining [165, 166, 167], but not generally applied in time series data processing. In this case multiple filter banks from multiple mother wavelets were employed. Each mother wavelet was used to decompose data to provide a set of filter banks, also known as packets and then a GA was used to evaluate a subset of the filters specified in each chromosome (
There are many types of Evolutionary Computations (EC). Each has strengths and weaknesses. Darwin's theorem influenced all of them. A brief introduction and history to GAs is provided and explains the differences between GA and other numerical methods. The no free lunch [138] theorem states there is not a universal algorithm to derive the global optimal solutions for all problems. Therefore, only a few ECs are investigated and explained in detail.
Evolutionary computation is a research domain, where computer algorithms are inspired by principles of natural evolution. Evolutionary computation has sub areas: evolutionary programming, evolution strategies, genetic algorithms, and genetic programming. All may be characterized as non-linear, stochastic search methods based on the mechanics of natural selection and natural genetics [138]. Survival of the fittest among the entire population provides the step forward to find the “optimal solution” from one generation to another. The three main mechanisms that drive evolution forward are reproduction, inheritance with variation, and natural selection. Usually, GAs use crossover and mutation to explore the landscape. Evolutionary Programming (EP) usually uses only mutation and emphasizes the evolution of behaviors. Evolution Strategies (ES) are based on adaptation and evolution, and traditionally used only mutation for variation. The selection process is often deterministic and based on fitness ranking and not on the evaluation value. In general, GAs work on a population of chromosomes. Each chromosome is represented by a string of genes, where each gene represents a variable in the evaluated chromosome. Based on chromosomes' evaluations, offspring are produced using crossover and mutation processes.
Evolution Strategies often employ more advanced mutation operators and often use smaller population sizes than GAs. Their solution representation and selection operators can be different from GAs. The best individuals from the population is allowed to produce offspring, which will represent the next population. The offspring are mutated and evaluated before they become the next generation. In some cases the new offspring will be combined with the previous population. The general principle of ES is as follows:
Genetic Programming (GP) is often used to evolve computer programs and sometimes to derive mathematical equations to represent a solution to a problem. GP is able to model dynamic systems using evolved equations which best represent a solution to a system. The problem representation is encoded as a tree structure, where branches are evaluated in a specific order. Evolutionary programming (EP) is similar to GP, but the structure of the program is fixed, similar to GA. N numbers of genes can be represented where in every generation some genes are mutated and represent different operations. The data flow is similar to GA, but the new population is the mutated old population.
The search for an optimal solution is desired in every challenge. Non-linear, stochastic search, based on the mechanics of natural selection are what characterize Genetic Algorithms [139]. They are a particular evolutionary algorithm class that uses biology inspired techniques such as mutation, selection, inheritance and crossover (recombination). Survival of the fittest among the entire population with some mutation provides the step forward towards the optimal solution from one generation to another. A new set of artificial chromosomes or individuals (population) is created using some or all chromosomes in every generation. Some genes in these chromosomes are combined with new genes which were not explored before or were inherited from previous generations. This may be considered to correspond, to a son who looks like his father, but yet has minor variations which distinguish him and are carried on from previous generations, say his grandfather. Genetic Algorithms are randomized yet not simple random. They allow efficient exploitation of historical information combined with the new, and performance usually improves across generations.
The search for the best solution may not be possible mathematically and, or may cost unrealistic computation time and effort. There are three main types of search methods identified by the current literature: calculus-based, enumerative, and stochastic. Calculus-based methods have been explored widely and divide to two groups: indirect method, where search for local maxima is by solving a nonlinear set of equations directed from the gradient of the objective function, and direct method, where search uses a function likely to move in the direction related of the local gradient to find the local optima. The main disadvantage of calculus based methods is local scope. The optima they search are the best in the neighborhood of the current location, but they may not explore sufficiently; missing sharp events is probable. Based on the starting point, calculus based methods potentially miss the highest hill and focus on a local hill. In general, enumerative schemes evaluate every point in the space, one at a time, similarly to a brute force approach. This method is applicable when the search space is small, due to the lack of efficiency in large space problems. As researchers recognized the effectiveness of stochastic search over calculus-based and enumerative schemes, it became more popular. However, stochastic methods can also lack of efficiency in large space problems. Traditional schemes work well in narrow space problems, but not in broad spectrum problems.
Genetic algorithms are different in their way of solving a problem from traditional methods [139]. First, GAs do not work with the parameters themselves, but rather the coding of the parameter set. Second, the search is done from a population point of view, and not from a single point; multiple searches are performed in parallel. Third, GAs use an evaluation function in the form of payoff, considering the problem as a black box, and do not use derivatives or mathematical functions. Advantageously, they need make no assumptions of this sort. Fourth, GAs use probabilistic transition rules and not deterministic rules to direct the search. Considering the black box problem, GAs will select some inputs and acquire the evaluated output. New inputs will be based on the probability of having better performance from the previous runs' evaluations until a certain degree of satisfaction is achieved. Crossover allows useful combinations to generate better results based on the previous results. Therefore, genes from the above average ancestors will be used more frequently, similar to Darwin's evolution scheme.
The initial population is often generated randomly; each individual is generated with random genes. Often, the initial population is a combination of two initial populations after evaluating each individual; the best from both are combined into one population. Therefore, the initial population starts with better individuals, which can aid in the reduction of computation time. In some GAs, not all the individuals within the population are mated, only the more fit individuals are selected. There can also be different restrictions depending on the GA type and close “cousins”, very similar individuals, are not allowed to mate. This is called incest prevention [147]. The individuals who mate produce offspring. These offspring may replace some or all of the population if they are more fit.
Each offspring is evaluated using a fitness function. It is very common to use fitness proportional reproduction, where each potential parent is compared to the population average, and this determines the expected number of offspring. The best individuals from the population are selected to continue into the next generation, whereas the others leave.
Given a problem with no prior knowledge, the landscape of all possible solutions may be defined as a finite space. Wolpert and Macready [138] showed that over the space of all possible search problems, all search algorithms have the same average performance. If algorithm A outperforms algorithm B on some set of problems, on all the other problems, the opposite will be true. In other words, there is no universally superior search algorithm.
In general, as the complexity level increases, a function's ability to provide a solution to a large problem decreases. Hill Climber (HC) functions are able to solve large problems where the complexity level is low and the landscape is defined with one high peak. At the other complexity end, Random Search (RS) and Exhaustive Enumeration (EE) are able to solve any complex problems, where the landscape is very complex with many peaks, but require small problem size. For intermediate levels of landscape complexity, algorithms like Iterated Hill Climbers (IHC), genetic algorithms (GA) and simulated annealing (SA) will each have a niche where they are able to solve the largest problems.
If the landscape complexity is unknown, the only way to understand its complexity is to perform exploration. For example, when one plays slot machines and wants to know which machines provides better outcomes; exploration is needed. Playing with each machine provides exploration of the landscape. With further exploration, patterns are recognized. One machine may provide better outcomes than others. One would like to have a strategy that exploits this emerging discovery. The optimal strategy for this, “k armed bandits' problem,” is known to allocate trials to machines in exponential fashion, where the exponent is the estimate of each machine's payoff relative to the population average. As a result, some machines will be played more often than others. Furthermore, the process is adaptive: one big successful outcome from another machine may result in a shift the exploration pattern.
Much work has been done to develop superior algorithms, yet the no free lunch theorem states there is no superior algorithm for all problems. However, based on a specific problem, or class of problems, and known conditions a good algorithm can be developed and perform better than others. Previous work had produced an EC with strong properties for subset selection tasks of varying complexity [148] It used Eshelman's CHC algorithm [147]. CHC GA is non-traditional GA, where the offspring replace inferior parents (cross-generational rank selection). A superior individual (chromosome) can survive across generations. Different crossovers were investigated by Mathias et al. [148], and Schaffer et al. [149] to provide an effective subset selection GA, and they found genetic respect to be the most important characteristic. Respect means that genes that are common between parents should be passed on to the offspring [151]. This concept was tested to learn how much respect is sufficient to provide the balance between exploration and exploitation. The final crossover that is used by the algorithm is called the mix and match crossover (MMX) SubSet Selection (SSS), where the size of each chromosome plays a rule in the desired solution.
The initial CHC (Cross generation rank selection, Half uniform crossover, Cataclysmic mutation) population is generated randomly using a uniform distribution. In CHC, two initial populations are produced and the chromosomes are evaluated, and the more fit chromosomes from both populations are selected to become the next population. For all subsequent generations, the pairs of parents (randomly mated) produce two offspring and the selection operator produces the next parent generation by taking the best from the combined parents and offspring using simple deterministic ranking. Each parent chromosome has exactly one mating opportunity each generation, and the resulting offspring replace inferior parents. Mates are randomly selected, but limited due to an incest prevention operator applied before the offspring reproduction crossover operator. There is no mutation performed in the “inner loop.” Only when it becomes clear that further crossovers are unlikely to advance the search, a soft restart is performed, using mutation to introduce substantial new diversity, but also retaining the best individual chromosome in the population. The responsibility for offspring production belongs to the crossover operator.
Crossover operator MMX is similar to RRR [147]. The basic MMX employs negative respect (i.e. no mutations). The MMX_SSS crossover operator consists of the MMX_0.85 crossover combined with a Subset Size (SSS) gene [148], which encodes the number of evaluated genes out of the total number of genes included in the chromosome. The parameter (0.85) in the operator name encodes the extent of negative respect (i.e. 15% mutations to the parental “unique” genes can occur).
SSS is a secondary mandate to the selection function. It is employed only if two chromosomes have the same value for classification accuracy (hierarchical selection with a two-dimensional fitness vector); then the smaller chromosome is ranked more fit. The offspring SSS gene is generated using Blend Crossover (BLX) [148, 151]. The SSS gene for each offspring is drawn uniformly randomly from an interval defined by the SSS genes in the parents and their fitness, the interval is first set to that bounded by the parental values, and then extended by in the direction of the more-fit parent.
See, e.g., U.S. Patent and U.S. Pat. Nos. 8,842,136, 8,849,575, 8,849,629, 8,855,980, 8,856,716, 8,862,627, 8,903,997, 8,913,839, 8,922,856, 8,923,981, 8,938,113, 8,945,875, 8,990,688, 8,995,074, 9,002,682, 9,009,670, 9,015,093, 9,015,145, 9,017,691, 9,022,564, 9,023,984, 9,029,413, 9,047,272, 9,047,353, 9,051,379, 9,053,416, 9,063,139, 9,121,801, 9,148,839, 9,164,481, 9,168,290, 9,171,250, 9,189,733, 9,193,442, 9,195,949, 9,213,990, 9,218,181, 9,223,569, 9,242,095, 9,256,701, 9,258,199, 9,259,579, 9,287,939, 9,289,471, 9,317,626, 9,317,740, 9,321,544, 9,323,890, 9,325,348, RE45660, 20160152438, 20160136431, 20160073313, 20160041174, 20160033622, 20160024156, 20150046181, 20150046216, 20150058061, 20150073495, 20150081077, 20150081324, 20150081911, 20150086087, 20150087589, 20150095146, 20150103681, 20150105086, 20150106310, 20150106311, 20150106314, 20150106315, 20150106316, 20150112403, 20150112636, 20150118158, 20150127066, 20150133306, 20150133307, 20150134315, 20150139977, 20150161629, 20150170052, 20150174408, 20150181822, 20150206214, 20150216426, 20150234976, 20150244946, 20150261926, 20150288573, 20150289210, 20150320316, 20150353880, 20150355459, 20150356350, 20150363108, 20150363193, 20150363194, 20150141863, 20150106069, 20150094012, 20150093037, 20150088024, 20150068069, 20150065835, 20150051083, 20150032015, 20150012256, 20140371834, 20140364721, 20140351188, 20140344013, 20140343396, 20140342703, 20140325019, 20140321756, 20140316221, 20140315576, 20140309959, 20140308930, 20140289172, 20140276191, 20140276121, 20140249429, 20140236530, 20140235965, 20140235201, 20140229409, 20140219566, 20140213909, 20140204700, 20140201126, 20140200575, 20140200572, 20140200571, 20140200471, 20140200430, 20140200429, 20140200428, 20140193087, 20140180049, 20140173452, 20140169686, 20140163425, 20140143251, 20140143064, 20140127672, 20140114165, 20140104059, 20140089241, 20140088415, 20140079297, 20140077946, 20140074564, 20140074180, 20140074179, 20140067485, 20140067484, 20140067470, 20140067463, 20140066738, 20140052379, 20140025304, 20140005743, 20130338530, 20130338496, 20130338468, 20130336594, 20130317392, 20130314694, 20130303941, 20130303119, 20130301889, 20130273968, 20130269376, 20130252604, 20130236067, 20130231574, 20130218156, 20130214943, 20130211291, 20130202177, 20130197380, 20130191090, 20130189977, 20130184603, 20130184553, 20130184538, 20130178730, 20130173194, 20130172691, 20130151447, 20130135008, 20130123684, 20130123666, 20130109995, 20130096394, 20130090266, 20130090265, 20130090247, 20130079002, 20130077891, 20130077843, 20130073981, 20130073490, 20130071837, 20130063613, 20130028052, 20130011062, 20130009783, 20120330170, 20120330109, 20120303560, 20120303504, 20120290505, 20120274937, 20120265350, 20120257046, 20120245481, 20120214510, 20120209798, 20120197831, 20120190404, 20120173154, 20120172746, 20120172743, 20120169053, 20120158633, 20120148157, 20120148149, 20120143382, 20120143078, 20120123232, 20120114249, 20120109653, 20120109612, 20120095352, 20120092157, 20120092156, 20120089046, 20120083246, 20120073825, 20120066259, 20120066217, 20120053441, 20120050750, 20120041608, 20120041330, 20120036016, 20120010867, 20120004854, 20120004564, 20110313285, 20110285982, 20110280447, 20110275364, 20110268328, 20110263958, 20110261178, 20110236922, 20110181422, 20110172930, 20110167110, 20110164783, 20110156896, 20110150253, 20110144519, 20110144065, 20110135166, 20110131162, 20110131041, 20110115624, 20110112426, 20110096144, 20110066404, 20110047105, 20110034967, 20110034824, 20110026832, 20110004513, 20110004415, 20100317420, 20100316283, 20100293115, 20100292968, 20100272340, 20100246544, 20100235285, 20100217145, 20100214545, 20100204540, 20100198098, 20100184702, 20100179400, 20100174271, 20100168836, 20100161654, 20100159945, 20100152905, 20100138026, 20100130189, 20100119128, 20100111396, 20100106458, 20100106269, 20100102825, 20100076642, 20100066540, 20100049369, 20100048242, 20100046842, 20100045465, 20100041365, 20100030485, 20100030484, 20100030038, 20100027892, 20100027846, 20100027845, 20100027469, 20100027431, 20100026799, 20100023307, 20100022855, 20100020961, 20100020208, 20100018718, 20100016687, 20100014718, 20100010681, 20100010503, 20100010488, 20100010368, 20100010355, 20100010332, 20100010331, 20100010324, 20090318779, 20090316988, 20090313041, 20090312819, 20090307164, 20090299162, 20090288152, 20090288140, 20090286512, 20090286509, 20090285166, 20090271342, 20090259537, 20090259534, 20090259533, 20090240366, 20090231173, 20090228408, 20090227876, 20090222108, 20090216133, 20090204341, 20090204029, 20090203981, 20090182287, 20090177420, 20090171740, 20090169075, 20090118637, 20090083010, 20090062684, 20090062635, 20090055147, 20090043542, 20090043541, 20090043525, 20090043182, 20090043181, 20090036758, 20090024549, 20090018891, 20090012766, 20080294019, 20080292146, 20080270328, 20080265130, 20080263323, 20080256069, 20080247598, 20080236275, 20080235165, 20080222734, 20080195261, 20080194996, 20080194946, 20080175480, 20080162487, 20080157510, 20080152217, 20080147441, 20080147440, 20080147438, 20080146334, 20080144944, 20080142713, 20080114564, 20080071136, 20080065291, 20080058668, 20080051660, 20080036580, 20080036187, 20080033316, 20080027841, 20080027769, 20080021342, 20080021336, 20080015871, 20080013747, 20080004904, 20080001735, 20070286336, 20070276279, 20070276270, 20070273504, 20070265808, 20070265806, 20070265533, 20070262574, 20070260656, 20070260427, 20070260425, 20070258329, 20070256432, 20070230795, 20070219749, 20070213786, 20070193811, 20070175998, 20070167846, 20070162992, 20070162189, 20070162084, 20070160973, 20070156317, 20070154099, 20070154079, 20070154078, 20070154063, 20070152433, 20070150021, 20070140551, 20070135984, 20070087756, 20070086624, 20070067003, 20070054347, 20070054266, 20070053513, 20070035114, 20070025597, 20070016476, 20060253781, 20060253258, 20060251293, 20060247536, 20060244246, 20060229822, 20060208169, 20060200260, 20060200259, 20060200258, 20060200253, 20060184473, 20060167784, 20060155398, 20060123363, 20060120584, 20060106797, 20060101017, 20060084115, 20060059028, 20060020597, 20060015497, 20060015496, 20060015495, 20060015494, 20060015492, 20050286179, 20050272110, 20050267911, 20050248136, 20050246314, 20050203434, 20050203360, 20050197590, 20050196047, 20050183958, 20050158736, 20050156775, 20050144284, 20050131660, 20050131607, 20050119454, 20050114078, 20050102246, 20050089923, 20050079524, 20050076190, 20050075846, 20050069162, 20050046584, 20050027457, 20050026199, 20050021101, 20050017488, 20050008179, 20040243567, 20040230131, 20040229210, 20040225649, 20040225629, 20040207548, 20040138578, 20040133355, 20040129478, 20040125148, 20040125133, 20040125121, 20040068199, 20040045030, 20040019470, 20030228565, 20030217047, 20030209893, 20030208451, 20030200189, 20030176656, 20030135109, 20030101164, 20030086593, 20030081836, 20030061228, 20030059837, 20030055799, 20030036835, 20020194159, 20020186875, 20020176624, 20020173936, 20020165854, 20020151992, 20020123975, 20020082756, 20020059022, 20020054694, 20020015532, 20020009756, 20010028743, U.S. Pat. Nos. 9,037,223, 9,033,893, 9,028,405, 9,019,819, 9,002,483, 8,996,442, 8,990,740, 8,976,856, 8,972,861, 8,965,044, 8,948,442, 8,942,180, 8,923,958, 8,918,169, 8,897,869, 8,897,586, 8,892,413, 8,886,301, 8,882,765, 8,880,158, 8,874,477, 8,874,203, 8,873,853, 8,873,813, 8,868,172, 8,866,936, 8,866,322, 8,861,799, 8,860,793, 8,855,775, 8,855,372, 8,855,011, 8,850,048, 8,849,737, 8,849,390, 8,838,510, 8,831,705, 8,826,199, 8,818,778, 8,818,404, 8,811,977, 8,810,796, 8,801,610, 8,797,550, 8,797,448, 8,792,974, 8,786,624, 8,781,597, 8,775,143, 8,775,134, 8,764,661, 8,764,651, 8,762,065, 8,761,903, 8,761,893, 8,761,051, 8,755,940, 8,755,916, 8,755,837, 8,750,971, 8,747,336, 8,747,315, 8,744,607, 8,744,557, 8,743,776, 8,725,667, 8,725,507, 8,725,243, 8,714,983, 8,713,025, 8,712,507, 8,702,629, 8,684,900, 8,680,991, 8,679,018, 8,678,943, 8,677,505, 8,659,697, 8,657,745, 8,649,565, 8,648,959, 8,645,832, 8,638,655, 8,635,051, 8,632,469, 8,626,274, 8,620,660, 8,611,692, 8,606,418, 8,606,021, 8,605,970, 8,600,830, 8,599,266, 8,595,164, 8,594,811, 8,588,933, 8,583,263, 8,582,916, 8,577,822, 8,577,451, 8,576,693, 8,560,134, 8,559,645, 8,547,824, 8,543,199, 8,536,133, 8,531,291, 8,527,324, 8,525,687, 8,525,673, RE44460, 8,522,312, 8,520,979, 8,516,266, 8,515,126, 8,503,791, 8,489,247, 8,488,863, 8,486,690, 8,478,394, 8,469,886, 8,467,884, 8,467,874, 8,467,611, 8,463,441, 8,461,988, 8,449,471, 8,437,998, 8,437,844, 8,437,223, 8,435,179, 8,435,167, 8,433,101, 8,428,925, 8,427,649, 8,416,710, 8,411,910, 8,406,867, 8,406,522, 8,406,115, 8,397,204, 8,396,582, 8,391,963, 8,388,604, 8,374,696, 8,374,667, 8,374,414, 8,369,967, 8,366,707, 8,364,136, 8,355,579, 8,351,321, 8,331,228, 8,323,188, 8,320,217, 8,315,150, 8,307,273, 8,305,436, 8,301,406, 8,295,934, 8,290,561, 8,282,549, 8,271,412, 8,265,725, 8,257,259, 8,253,824, 8,251,906, 8,244,475, 8,233,958, 8,226,561, 8,216,139, 8,213,399, 8,213,398, 8,209,745, 8,208,697, 8,204,697, 8,200,506, 8,199,632, 8,194,986, 8,194,938, 8,190,543, 8,190,194, 8,185,194, 8,183,062, 8,179,847, 8,174,956, 8,170,335, 8,170,334, 8,170,333, 8,165,916, 8,165,661, 8,164,345, 8,149,649, 8,126,664, 8,121,823, 8,121,046, 8,114,143, 8,108,036, 8,107,726, 8,103,333, 8,099,161, 8,098,938, 8,094,551, 8,089,853, 8,086,294, 8,086,017, 8,082,353, 8,082,032, 8,078,552, 8,078,274, 8,077,958, 8,068,894, 8,055,667, 8,046,107, 8,041,651, 8,041,124, 8,036,736, 8,036,442, 8,036,265, 8,032,477, 8,031,060, 8,023,710, 8,015,128, 8,005,631, 8,005,524, 7,996,762, 7,995,454, 7,987,003, 7,983,817, 7,983,141, 7,981,399, 7,974,714, 7,966,078, 7,957,265, 7,936,662, RE42236, 7,912,734, 7,912,138, 7,904,187, 7,903,617, 7,887,089, 7,881,181, 7,881,180, 7,872,985, 7,855,977, 7,831,358, 7,823,058, 7,819,003, 7,818,053, 7,813,822, RE41771, 7,805,386, 7,788,212, 7,777,743, 7,773,537, 7,769,513, 7,768,380, 7,756,060, 7,747,390, 7,747,325, 7,742,806, 7,733,224, 7,730,063, 7,716,148, 7,712,898, 7,710,828, 7,706,349, 7,702,555, 7,702,185, 7,697,792, 7,697,453, 7,693,683, 7,676,263, 7,676,062, 7,675,843, 7,672,219, 7,668,697, 7,663,502, 7,662,785, 7,660,437, 7,657,299, 7,655,895, 7,649,160, 7,636,700, 7,630,757, 7,624,076, 7,620,527, 7,606,790, 7,604,956, 7,599,759, 7,596,470, 7,596,242, 7,590,589, 7,590,510, 7,587,069, 7,584,075, 7,581,434, 7,575,171, 7,558,622, 7,539,533, 7,539,532, 7,536,064, 7,535,822, 7,533,006, 7,526,461, 7,519,476, 7,511,833, 7,502,677, 7,483,868, 7,477,758, 7,460,903, 7,454,244, 7,451,005, 7,430,483, 7,428,323, 7,426,499, 7,415,126, 7,409,303, 7,408,486, 7,407,029, 7,403,820, 7,401,807, 7,401,057, 7,395,250, 7,392,143, 7,385,300, 7,383,237, 7,379,568, 7,376,553, 7,366,719, 7,333,851, 7,324,851, 7,324,036, 7,310,522, 7,308,126, 7,295,608, 7,293,002, 7,286,964, 7,277,758, 7,270,733, 7,243,945, 7,242,984, 7,233,882, 7,231,254, 7,228,238, 7,209,787, 7,194,143, 7,190,149, 7,180,943, 7,164,117, 7,162,076, 7,149,320, 7,147,246, 7,085,401, 7,082,572, 7,051,017, 7,039,654, 7,028,015, 7,016,885, 7,007,035, 7,006,881, 7,003,403, 6,996,549, 6,988,093, 6,970,587, 6,886,008, 6,885,975, 6,882,992, 6,879,729, 6,865,492, 6,862,710, 6,850,252, 6,826,428, 6,826,300, 6,801,645, 6,789,054, 6,763,322, 6,763,128, 6,757,602, 6,757,415, 6,708,163, 6,697,661, 6,678,548, 6,675,164, 6,658,287, 6,650,779, 6,650,766, 6,640,145, 6,601,051, 6,560,542, 6,556,699, 6,549,804, 6,535,644, 6,529,809, 6,510,406, 6,459,973, 6,452,870, 6,445,988, 6,434,583, 6,400,996, 6,397,136, 6,389,157, 6,377,306, 6,363,350, 6,334,219, 6,272,479, 6,205,236, 6,167,155, 6,167,146, 6,154,705, 6,137,898, 6,128,346, 6,121,969, 6,115,488, 6,098,051, 6,091,841, 6,012,046, 5,995,868, 5,978,788, 5,963,929, 5,940,825, 5,847,952, 5,845,266, 5,815,608, 5,815,198, 5,745,382, 5,649,065, 5,602,964, each of which is expressly incorporated herein by reference in its entirety. These references disclose, for example, complementary technologies, aspects and details of implementation, applications, and the like.
The present technology provides a non-invasive and continuous system and method for monitoring of cardiac output from the human being. The technology is based on using high-sensitivity accelerometers to record chest wall motions. Specific filtering algorithms are utilized to identify both ventricular contraction time and stroke volume from such chest wall motions.
Advantageously, the system and method provide usable measurements of cardiac output for individuals who are upright, during locomotion. Therefore, the techniques are not limited to supine subjects or to subjects while holding their breath. Thus, this technology may find use in sports monitoring, and ambulatory patient monitoring. Further, the technology may be integrated within an implantable medical device (though likely requiring special calibration), and may be used to adjust a pacemaker, for example. In such a device, the cardiac output may be used as a feedback variable, permitting cardiac output to automatically respond to patient need.
In a sports medicine implementation, an athlete may be provided with real time feedback to optimize performance based on available blood flow, and perhaps predictive feedback during endurance sports to balance oxidative and glycolytic metabolism. The feedback may be presented through a watch display, audio feedback, or the like. After a sporting event or practice, the cardiac output measurements may be compared with athletic performance, to assist the athlete in achieving higher or better performance. The monitor can also be used for non competitive sports, and the cardiac output is measured to monitor health and optimize a workout. For impaired patients, the cardiac output monitor may be provided as an indicator to provide an assessment, and help determine an upper limit on physical exertion.
The technology can be implemented as a low power, portable or wearable, battery operated device. Numerous applications in wellness care, medical care, exercise monitoring, sleep monitoring, and sport science are proposed. The technology may also be applicable to veterinary science.
Different investigators have attempted to correlate specific chest acceleration signal components with the heart mechanical activity related to valves functions and volume movement. To date, all published studies have focused on analyzing the entire signal, that is, every detail of the recording throughout the cardiac cycle. Attempting to interpret all of the detailed information in a chest wall acceleration signal is problematic due to differences between subjects. However, the inward deviation of the chest wall associated with ventricular contraction which happens during the first heart sound is an aspect of the chest wall acceleration signal which is both uniform and consistent. A specific filter set was developed that is capable of providing an accurate estimate of stroke volume.
Hardware and signal processing techniques suitable for extracting both ventricular contraction timing and stroke volume estimation in the presence of breathing and when a person is moving (e.g. exercising) were developed. A 24-bit ADC is used to record chest acceleration where high amplitude impact noise does not saturate the input signal. Stroke volume and cardiac output estimates obtained using this new technology have been compared to simultaneous stroke volume and cardiac output measurements obtained with FDA approved electrical impedance technology (NICOM by Cheetah Medical) and a good correlation to this standard measurement technique, of approximately 90%, was demonstrated.
The principle challenges in achieving an accurate and reproducible estimate of stroke volume from chest wall motion is obtaining accurate timing of when the ventricle is contracting so that the signal analysis algorithm can be focused on only the relevant portion of the recorded signal, and removal of the extraneous noise generated through breathing, internal body organ movements, and whole body motion. The latter three tasks can be accomplished by appropriate filtering. Wavelet packet analysis is used for extracting information from noise in non-stationary time series data. A usual approach is to use Wavelet analysis to denoise the data and to separate observed components which subsequently allows decomposition, thresh holding, and reconstruction [176, 190, 191]. Moreover, in wavelet analysis, the mother wavelet is typically chosen based on the experience of the designer, and his sense of the ability of the mother wavelet to match certain features of the time series data stream. In some cases, just decomposition (without reconstruction) is performed to capture the magnitude of some specific feature of the signal [115].
Rather than a-priori selecting a mother wavelet for analyzing chest wall motion signals, multiple decompositions were performed based on a variety of different mother wavelets, producing an extremely large number of potential filter packets. A unique subset selection genetic algorithm (SSS-GA) was then used to isolate an optimal filter set that allows extraction of stroke volume output from acceleration recordings obtained from the chest wall during exercise using a relatively compact filter set. Previous investigations [147, 148, 149] demonstrated that a preferred strategy in this approach is to incorporate an index representation (genes are the indexes of the features to select from a possibly large pool of features), with a special subset selection crossover operator. This approach was utilized on the continuous acceleration signal to provide a continuous stroke volume assessment. Previously the SSS-GA algorithm was used with a classifier. However, the SSS-GA algorithm is now used to provide a continuous estimate. The evaluation function utilizes multiple linear regressions. The coefficient of determination is limited to two significant figures in order to provide sufficient quantization noise to promote algorithm convergence.
Four healthy men between 22-24 years in age were used as subjects. BMI ranged between 21 and 30 Kg/m2, and heights ranged from 173 to 188 cm. One subject exercised regularly, while the others did not. Two subjects were Caucasian, two Asian. Cardiac output was recorded via electrical impedance spectroscopy (NICOM, Cheetah Medical, Inc.) using four electrode sets; two placed on the lower thorax and two on the upper thorax, and was set up to sample stroke volume, heart rate, and cardiac output every thirty seconds. Three ECG electrodes were used to obtain the second ECG lead, with electrodes placed on upper thorax on the right, and two on the lower thorax. Chest accelerations were recorded on the sternum above the xiphoid using a 2G accelerometer (Silicon Designs Model 1221), with analog filtering used to capture motion in the 8 Hz to 370 Hz to reduce high and low frequencies noises before digitalizing the input acceleration signal. A BioPack MP-150 data acquisition system was used to collect and digitalize (at 2000 Hz) the analog data. A twenty pole low pass digital filter at 50 Hz was used to isolate the frequencies between 8 Hz and 50 Hz to focus on the heart frequency spectrum and remove high frequency components. Following low pass filtering the data were decimated by factor of 20 to set the Nyquist frequency to 50 Hz.
Measurements were taken while subjects were at rest and during exercise. Data collection was started while the subject was at rest, in supine position, for 240 seconds, and then upright for 210 seconds. Five exercise sessions were started consisting of an exercise period for 150 seconds, pedaling at 100 cycles per minute, followed by a 270 second resting period.
A strap was wrapped around the subject with sufficient force (3N) to hold the accelerometer in place while exercising without creating difficulty to breathe. The NICOM provides thirty-second averages of stroke volume, and so wavelet decomposition was performed on each thirty seconds of recoded acceleration data. Eighty five 30s averaged measurements were taken sequentially using the NICOM, the ECG, and chest accelerations from each subject.
A filter set was identified using SSS-GA. Four mother wavelets were utilized in the design process, Daubechies (db); Symlets (sym); Coiflets (coif); BiorSplines (bior); ReverseBior (rbio).
Table 1 shows the best filter set identified to date for extracting stroke volume from chest wall acceleration data. This is a 28 filter set which utilizes packets arising from all four mother wavelets combined with Body Surface Area (BSA) of the subject, and a constant.
A multiple linear regression evaluation (using equation 1 (below) and table 2 parameters, where, W—weight, MW—mother wavelet, P—packet) was used to compare NICOM obtained stroke volume estimates to the chest wall based stroke volume estimates:
SV=4.15·BSA−14.5+W1·MW1(P1)+W2·MW2(P2)+ . . . +Wi·MWi(Pi)
Accurate ventricular contraction times are critical to the performance of the above described algorithm. In initial algorithm development, ECG recordings were used to identify the point of ventricular contraction. In order to eliminate the need for the ECG, a procedure was developed to extract left ventricular contraction times directly from the acceleration signal.
Ventricular contraction timing is obtained using two algorithms, an adaptive heart rate detection algorithm, and a contraction location detection algorithm. Combined, these two algorithms provide a high correlation to the ventricular contraction time-point based on ECG recordings. In order to develop the adaptive algorithm pair, chest wall acceleration signals were concatenated from seven test subjects with different BSA and BMI. Focus is on contraction detection, since missing a heartbeat while calculating cardiac output is not permitted.
The heart rate detection algorithm isolates the first major acceleration deviation in the acceleration signal from the second major deviation (in terms of heart sounds, this is equivalent to finding the time point of the first heart sound without confusing with the second heart sound). In addition, this algorithm provides an estimate of subject heart rate (number of heartbeats per minute).
The ventricular contraction algorithm is based on a Continuous Wavelet Transform (CWT) approach, where the scaling of the mother wavelet is related to the measured hear heart rate. Filter shifting properties are used to isolate the first heart acceleration deviation (Contraction period). This algorithm provides a threshold window where only the positive data are analyzed and negative data rejected. Heart rate is measured by counting the number of threshold windows per minute, and is used to scale the CWT ‘db5’ mother wavelet accordingly every 60 seconds. The algorithm was tested on seven subjects between 22 and 60 years old, seated in an upright position, at rest, while breathing. The equation below is used to scale the CWT threshold function as a function of heart rate (HR). If the measured HR is greater than 240, CWT is set to be the maximum value the mother wavelet is scaled to 20, otherwise the scaling equation is used.
The scaling algorithm proved to be able to adjust to a sudden new heart rate, as well as being able to adjust to different acceleration signals obtained from various subjects.
Following the CWT Heart rate/window threshold algorithm, a second algorithm is used to detect the heart contraction deviation location using the defined CWT threshold windows.
Originally, the data were collected and analyzed first with the CWT threshold function and then with the second algorithm to find the best filter set for identifying the contraction time within the CWT threshold window.
The best mode algorithm performs both rate and location detection at the same time. The CWT threshold function is computed while the contraction time location algorithm detection is also computed. When a CWT window identifies a threshold time segment (i.e. change from 0 to 1 and from 1 to 0) a search for global energy maxima is performed on the combined filter set for contraction time location.
Two different algorithms have been defined; one has better sensitivity while the second has better positive predictive values. The first solution consists of two filters combined by element multiplication. This solution has sensitivity of 0.986 and positive predictive values of 0.987 and is composed of element multiplication of ‘db8’ at scaling of eight and ‘Meyer’ at a scaling of twelve.
The alternative solution consists of three filters and provides better positive predictive values (PPV=0.991) than the first solution. However, it has lower sensitivity (0.983). This solution consists of ‘db13’ at a scaling of eight, ‘Meyer’ at a scaling of twelve, and ‘gaus2’ at a scaling of 106.
The three algorithms can be combined to provide stroke volume, heart rate, and cardiac output assessment using one accelerometer placed on the sternum. They are combined in the following order since each algorithm depends on the output of the previous:
1. CWT Threshold—detects a window around the first acceleration deviation.
2. CTLD—finer detection of the contraction time detection based on CWT Threshold window
3. SV calculation using multi-regression line based on the filters energy values at the contraction time given by the CTLD output function.
The first algorithm provides an adaptive threshold window to isolate the heart contraction phase from relaxation phase. The second algorithm defines the contraction time location at the adaptive threshold window of the second algorithm. Heart rate calculation provides the scaling function of the second algorithm. The third algorithm is applied on a continuous signal, where SV is calculated using the defined contraction time to extract the combined energy from the filter set. All are combined as the “Infrasonic stroke volume”.
An objective of the present technology is to provide accurate and reproducible cardiac output monitoring in a small, battery operated (i.e. portable) device, specifically, a device that will allow long term cardiac monitoring. By identifying a small set of filters which provide high correlation to a “gold standard” CO measure, it is possible to utilize a microcontroller device with relatively low levels of computational power to achieve SV and CO outputs in the available time (i.e. between two adjacent heartbeats, e.g. less than 250 ms for a heart rate of 4 Hz, or 250 bpm). For example, the ARM cortex M0+ processor, operating with a 3V battery, can provide 0.9 Dhrystone Million Instructions Per-Second (DMIPS), allowing the complete algorithm to finish before new data point is presented.
Thus, the present technology estimates SV from a seismocardiogram recording, which is obtained by recording chest wall acceleration, preferably at the xiphoid process [109, 192]. This approach involves performing multi-wavelet decompositions on the acceleration data to generate a large pool of features from which a genetic algorithm (GA) is used to select the best packet combination for predicting SV. In a sitting position, a patient's back is not against a firm surface, similar to standing, resulting in different chest motions than in the supine position. Moreover, the organs of the body are shifted towards the abdomen, changing the orientation and fluid motion in the body. The patient chest vibration is preferably measured while patient breathes normally.
Cardiac output may be extracted from the displacement signal while patient is in a seated position. A twenty pole digital low pass filter was used to reduce frequencies greater than 50 Hz. It is possible to utilize a 3-D accelerometer to define the chest angles relative to gravity, and optimize the cardiac output algorithm based on the two angles. One accelerometer detects the sternum angle relative to gravity in one direction, forward and backwards. The second accelerometer detects the relative chest angle to the left and right side of the body. The genetic algorithm is used to define the best packet set for the different positions. An adaptive system based on subject position will provide the opportunity to wear this device and compute subject daily routine.
It is therefore an object to provide a method for computing cardiac output, comprising: measuring chest wall movements of a subject; performing a wavelet transform on the measured movements; determining, from the wavelet transformed measured movements, at least a cardiac stroke volume, based on calibration data derived from a plurality of subjects; and outputting information selectively dependent on the determined cardiac stroke volume.
It is also an object to provide a method for computing cardiac stoke volume, comprising: measuring chest wall movements of a subject; performing a wavelet transform on the measured movements; determining, from the wavelet transformed measured movements, at least a cardiac stroke volume, based on calibration data derived from a plurality of subjects; and outputting information selectively dependent on the determined cardiac stroke volume.
The movements are detected as vibrations comprising 2-50 Hz, and more preferably 5-25 Hz. The measured movements may be detected with a position detector (e.g., interferometer) or accelerometer, for example.
The method typically also determines a heart contraction timing, which may be dependent on the measured movements, an ECG, a microphone listening for heart sounds, a microwave sensor, an echocardiogram sensor, etc. The system preferably also measures a heart rate, and this permits calculation of cardiac output as the product of heart rate and stroke volume.
The wavelet transform may be based on one or more optimal mother wavelets and associated parameters, determined using a genetic algorithm. The genetic algorithm may also optimize a polynomial function to process the wavelet transformed information to determine stroke volume. The wavelet transform may be optimized based on both low computational complexity and correlation with a benchmark, i.e., stroke volume or cardiac output from a verified source. The wavelet transform may also be derived from at least one of an evolutionary algorithm and a genetic algorithm, which optimizes correlation with a benchmark and computational complexity to produce a set of optimum mother wavelets and weights. The method may further comprise employing at least one of an evolutionary algorithm and a genetic algorithm to define an optimal wavelet transform, which optimizes correlation with a benchmark and computational complexity. Both the stroke volume and heart contraction timing may be determined based on the wavelet transformed measured movements, and determining a cardiac output, wherein the wavelet transform is optimized based on both low computational complexity and correlation with a cardiac output benchmark for the plurality of subjects. The wavelet transform may be derived from at least one of an evolutionary algorithm and a genetic algorithm, which defines an optimum set of mother wavelets and weights. The optimum set of mother wavelets may comprise at least two different types of wavelets. The wavelet transform may comprise at least two different types of mother wavelets, in different decomposition paths. The wavelet transform may comprise a plurality of different mother wavelet types, in a plurality of decomposition paths, in a plurality of different filters at different frequencies.
The method may further comprise applying a plurality of different filters to the measured movements.
The movements may be measured on the sternum, and preferably the xiphoid process.
The method may calculate a non-normalized cardiac output value (e.g., a distorted cardiac output value that can be corrected based on subject parameters to yield a normalized cardiac output value). The non-normalized cardiac output value may be stored, communicated over a communication link, and normalized to produce a cardiac output value normalized for at least one body characteristic of the subject. The communication link may comprise Bluetooth.
The non-normalized cardiac output value may be communicated over the Bluetooth communication link to a smartphone, and the smartphone may normalize the non-normalized cardiac output value to produce the cardiac output value normalized for the at least one body characteristic of the subject.
The method may also determine a cardiac ejection fraction, a variability of a stroke volume, a variability of a cardiac output, or other parameters.
The method may employ a system comprising a housing 1 containing at least: a movement sensor, e.g., transducer 7; a microcontroller, e.g., automated processor 2; a memory 4; a power source, e.g. self-contained power source 5; and a communication port 3, wherein the measured movements are determined by the movement sensor, the stroke volume is determined by the microcontroller powered by the power source, stored in the memory, and communicated through the communication port. An acceleration sensor 6, discussed below, may also be provided in the housing 1. The housing may have a volume of less than about 4 cubic inches, e.g., <2″ diameter and ¾″ thick. The housing may be configured to be wearable by a human affixed to the sternum. The microcontroller may have an average power consumption while determining stroke volume on a beat-by-beat basis of less than about 300 mW, and be powered by a rechargeable battery. The battery may be, for example, a 2,400 mW-Hr battery to provide for 8 hours of continuous use.
Another object provides a system for computing cardiac output, comprising: a transducer configured to sense chest wall movements of a subject; at least one automated processor, configured to: perform a wavelet transform on the measured movements based on at least one predefined mother wavelet and at least one set of weights; and determine, from the wavelet transformed measured chest wall movements, at least a cardiac stroke volume, based on calibration data derived from a plurality of humans; and at least one of an output port and a memory, configured to receive information selectively dependent on the determined cardiac stroke volume. The system may further comprise a housing configured to be wearable by a human, containing at least: the transducer having an output corresponding to chest wall movement; a self-contained power source; and the at least one automated processor, comprising a microcontroller powered by the self-contained power source configured to process the transducer output on a beat-by-beat basis, to produce information adapted for estimating the cardiac stroke volume, to store the information in the memory, and to communicate the stored information through the communication port; wherein the communicated stored information is adapted to be processed by a remote system to calculate a cardiac output. An acceleration sensor may also be provided, configure to determine an acceleration vector of the housing, the acceleration sensor being distinct from the transducer, the microcontroller being further configured to determine an artifact condition based on the acceleration vector.
The method may further comprise measuring an acceleration vector of the housing through an accelerometer distinct from the movement sensor, and determining an artifact condition based on the measure acceleration. This is particularly helpful where the movement sensor for determining the stroke volume is readily saturated by significant body movements, and the subject is unconstrained and moving.
While it is preferred that the cardiac timing be determined from the same chest wall movements as are used to determine the stroke volume, the timing can also be determined using electrocardiogram input, phonocardiogram input, microwave, echocardiogram, or other types of sensors.
The method may further comprise receiving a height and a weight of the subject, and calculating a cardiac output based on at least the height, weight, stroke volume, and a heart rate.
The method may further comprise persistently storing in a memory a value calculated dependent on the stroke volume for each heartbeat.
The subject may be a human, mammal, or other animal. The chest wall movements of the subject may be measured during breathing and/or physical exertion. For quadrupeds, a distinct optimization would be required.
It is a further object to provide a system for computing cardiac output, comprising: a transducer configured to sense chest wall movements of a subject; an automated processor, configured to: perform a wavelet transform on the measured movements based on at least one predefined mother wavelet and at least one set of weights; and determine, from the wavelet transformed measured movements, at least a cardiac stroke volume, based on calibration data derived from a plurality of humans; and at least one of an output port and a memory, configured to receive information selectively dependent on the determined cardiac stroke volume. The automated processor may comprise a reduced instruction set computer (RISC) processor, e.g., an ARM processor.
It is a still further object to provide a method for analyzing a biological signal, comprising: receiving the biological signal as a series of digitized values representing a continuous process; performing a wavelet transform on the series of digitized values; calculating a polynomial function of the wavelet transformed series of digitized values; and outputting a calculation result of the polynomial function. The outputted calculation result may be determined substantially without inverting the wavelet transform. The wavelet transform may comprise a plurality of different wavelet waveforms and a plurality of different decomposition paths.
The wavelet transform and polynomial function may be together derived based on at least a genetic algorithm and at least one fitness criterion.
The continuous process may comprise, for example, electroencephalogram signals, or electromyogram signals.
A further object provides a method for analyzing a signal, comprising: receiving a series of digitized values representing a physical process; defining at least one fitness criterion for a wavelet transform on the series of digitized values comprising a convolution of a plurality of different mother wavelet waveforms in a plurality of different decomposition paths, the wavelet transform comprising a plurality of respective terms and at least one respective parameter applied to each respective term; optimizing the at least one respective parameter applied to each respective term of the wavelet transform, based on at least an iterative genetic algorithm and the defined at least one fitness criterion; and storing the optimized at least one respective parameter applied to each respective term of the wavelet transform. The plurality of respective terms of the wavelet transform and at least one respective parameter applied to each respective term may be together optimized based on the at least an iterative genetic algorithm and the at least one fitness criterion. The at least one respective parameter applied to each respective term of the wavelet transform may be calculated based on at least a correlation R2 of the wavelet transform with a reference at a respective stage of the iterative genetic algorithm, and a precision of the correlation R2 is sufficiently limited in order to increase a rate of convergence of the iterative genetic algorithm. The method may further comprise receiving a time-continuous biological signal; digitizing the biological signal to form the series of digitized values; and employing the stored optimized at least one respective parameter applied to each respective term of the wavelet transform to process the series of digitized values to determine a biological parameter, substantially without inverting the wavelet transform.
The biological signal may comprise at least one of an electroencephalogram signal, an electrocardiogram signal, an electromyogram signal, a phonocardiogram signal, a ballistocardiogram signal, an ultrasound signal, an x-ray signal, a magnetic resonance signal, a lung sound signal, and a bowel noise signal.
It is a still further object to provide a method for analyzing a signal, comprising: receiving a series of digitized values; performing a wavelet transform on the series of digitized values comprising a plurality of different wavelet waveforms and a plurality of different decomposition paths; calculating a polynomial function of the wavelet transformed series of digitized values; and outputting a calculation result of the polynomial function, wherein the wavelet transform and the polynomial function are together optimized based on at least a genetic algorithm and at least one fitness criterion.
The present technology provides a system and method for calculating cardiac output based on infrasonic chest wall movements, in an unconstrained subject. An accelerometer is provided to measure xiphoid process movements, which provide indication of both heart rate and stroke volume. While heart rate can also be obtained from other measures, the present technology permits (but does not mandate) use of a single transducer. The sensor data is processed by an algorithm that provides high correlation to standard measures of cardiac output, and has high self-consistency. To extract the two components of cardiac output, HR and SV, two different algorithms were developed. First, the HR algorithm was developed using a wavelet-based decomposition using a genetic algorithm to optimize the mother wavelet and associated parameters. The resulting algorithm determines both the heart rate and time of ventricular contraction (ejection). Second, the SV algorithm, synchronized by the HR algorithm, analyzes chest wall movement, e.g., velocity, to estimate the ejection volume. Together, these algorithms can execute on relatively low resource computational platform to provide stroke by stroke calculation of cardiac output in real time, i.e., the calculations are complete before the next heartbeat.
Cardiac Output is defined as the amount of blood the heart pumps per minute. At each heartbeat the heart contracts, the blood is pushed out from the left ventricle into the aorta. Due to this movement, chest volume is decreased, displacing the sternum to inwards. This displacement results in acceleration at the sternum. The accelerometer captures this acceleration, which is analyzed to calculate the chest displacement. In general, the displacement is equal to the double integral of acceleration.
This technology is applicable to medical, sports and fitness, and non-invasive monitoring environments. It is believed that the technology may also be used in veterinary environments, with the parameters of the algorithms recalculated depending on the species and other physical parameters.
The optimization of the parameters of the algorithms do not need to be replicated in the target monitoring device, but significant modifications of underlying presumptions would suggest reoptimization. Therefore, the target device need only receive the accelerometer or other infrasonic pickup device output, filter and preprocess the data, and execute the algorithm, which may be dependent on a subject's sex, body surface area, weight, or other readily ascertainable physical characteristics. While it is preferred that a single algorithm subject to these inputs be used, it is of course possible to provide a family of algorithms that are selected and employed dependent on the physical subject attributes and context. For example, the algorithm may differ for patients suffering from various heart diseases than for healthy subjects, e.g., mitral valve prolapse, where cardiac output as reflected in aortic flows may require correction of the stroke volume for reflux. Similarly, cardiomegaly may require use of corrections or a distinctly optimized algorithm.
It is also noted that, for any given subject, the target device may adaptively optimize its implementation to compute relative cardiac output, or a related measurement, though absent a calibration standard, absolute cardiac output calculation requires use of a verified algorithm.
A multiaxis accelerometer permits intrinsic determination of patient posture and physical activity, which can also be used as inputs to the algorithm.
The technology may also be integrated with other sensors, such as ECG, echocardiogram, microwave (radar) chest sensing, phonocardiogram, pulse oximeter, blood pressure, respiration sensor, peripheral vascular resistance (see Sharrock et al., U.S. Pat. Nos. 8,821,403; 7,727,157; 6,994,675), body fluid chemistry (e.g., saliva or sweat CO2 and pH), and other non-invasive, minimally invasive or invasive measurements.
The technology may be implemented in a miniature form factor, and for example provide a module that adheres to the chest wall. The module may comprise the entire system, i.e., housing, sensor, analog signal processing (if employed), microprocessor, data memory, program memory, power supply, user interface, and data communications, or merely the housing, sensor, signal processing, and communications (e.g., Bluetooth), without execution of the algorithm. In the latter case, the cardiac output may be determined by an associated computing device, such as a smartphone, which receives the sensor data through the communication interface, and provides a platform for execution of the algorithm, user interface, and remote data interface.
The wavelet transform is a popular analysis tool for non-stationary data, but in many cases, the choice of the mother wavelet and basis set remains uncertain, particularly when dealing with physiological data. Furthermore, the possibility exists for combining information from numerous mother wavelets so as to exploit different features from the data. However, the combinatorics become daunting given the large number of basis sets that can be utilized. Recent work in evolutionary computation has produced a subset selection genetic algorithm specifically aimed at the discovery of small, high-performance, subsets from among a large pool of candidates.
This algorithm may be applied to the task of locating subsets of packets from multiple mother wavelet decompositions to estimate cardiac output from chest wall motions while avoiding the computational cost of full signal reconstruction. A continuous assessment metric can be extracted from the wavelet coefficients, but the technology preferably achieves a dual-nature objective of high accuracy with small feature sets, imposing a need to restrict the sensitivity of the continuous accuracy metric in order to achieve the small subset size desired.
Pilot studies were conducted using a standard MEMS accelerometer (Kistler Model 8312A). Different recording locations on the chest wall were investigated as well as the required filtering and equipment necessary to accurately and reproducibly extract CO measurements. Various analog filters, digital filters, and basic mathematical analysis approaches for removing noise and recording artifacts from the acceleration signal were also investigated. These initial studies relied on integrating the acceleration signal to obtain a displacement signal from which SV was determined. Polynomial curve fit baseline subtraction was the initial approach used to remove slow trends associated with integration and breathing. Wavelet analysis was used to remove noise from the recorded signal. First and second correlation to a NICOM Bio-impedance device showed good CO correlation.
The first recordings were undertaken in order to test the possibility of capturing a reproducible signal and provided cardiac information. Similar to previous studies reported in the literature, the initial recordings were performed with the subject supine (lying down on their back) and holding their breath for a period of 20 seconds. Three different recording locations were selected based on cardiac recording techniques utilized by others. These locations were assessed to minimize noise and other artifacts from the cardiac signal. The signal than was analyzed to extract cardiac information. Filtering and polynomial fit base subtraction were used to analyze the recorded data and provided repetitive waveform. Polynomial fit base subtraction provided consistent results to estimated CO from the recorded signal.
The recording location of any physiological signal is essential component in signal fidelity and reproducibility. Numerous considerations need to be taken into account when selecting the recording location. Muscle, fat, bones and cartilage and personal comfort are some of the components which effect the decision for recording location. Skeletal muscle vibrations in the range of 8-150 Hz are produced when contraction occurs and so can contribute to background noise in the infrasonic frequency range. Fat may contribute to low frequency vibration and also isolate or reduce specific frequencies. Bones and cartilage, in general, will transfer most acoustic energy since they are relatively solid matter, however they have a very different acoustic impedance than soft tissue, so will reflect a large portion of acoustic energy arising in soft tissue. With respect to pericardial motion recording, the skeletal system plays a critical role as the chest wall must flex in order to permit chest wall motion recording. A rigid rib cage will severely limit that motion of the chest wall. Comfort is also important to patients and as recording accuracy. If patients are not comfortable, the device may not be placed correctly or will shift from the original location over time.
Recordings were taken during breath holding and regular breathing using a 2G Kistler accelerometer, with pre-amplification (Model 5210). Initial recording showed that the sternum location provided the most consistent measurement sites since there is typically little muscle or fat at this location. This location is also easy to find and is symmetric compare to the other locations. The Apex location and its distance from the chest walls vary from one person to another, depends on subject physicality. In general, the apex is about 0.53±0.53 cm from the inner wall of chest, and 2.76±0.80 cm from the chest surface while subject is in supine position [127]. Moreover, females may have difficulty to place a transducer at the apex location. The upper chest location, between ribs two and three, slightly left from the sternum, has substantial underlying skeletal Pectoralis Major muscle, which can disturb the recorded signal if these muscles contract. Also, one person may place the transducer at slightly different locations than another, similar to what can occur at the apex location. Measurements were taken from one individual and were analyzed resulting in the selection of the lower sternum location above the xiphoid as the optimal recording location. The signal may be amplified digitally by a gain of 100 and sampled at 2,000 Hz.
The recorded signal demonstrates significant higher frequency components and has an offset due to the capture of the earth's gravitational field. Therefore, two filters were used as a band pass filter to capture frequencies between 0.05 to 150 Hz.
Heart Physiology
The ECG signal starts with the P wave deflection, result associated with both atria contracting, and correspondingly to an outward deflection of the chest wall, at maximum peak location. Ventricular contraction follows and is represented by the negative going curve (inward movement of the chest wall). SV is calculated using the slope connecting these two maxima. Isometric contraction occurs and provides constant blood pressure for a short period of time, directly correlated to systolic blood pressure. Following the ECG T wave, the ventricles relaxes and eccentric contraction occurs. The ventricles start to refill while the blood moves from the aorta out to the rest of the body. The displacement signal does not fully agree with the seismocardiogramy signal. For example, the acceleration signal shows MC—mitral valve close and AO—Aorta valve open deflections. The displacement signal does not show those events since the heart muscle is in isometric contraction during this period, resulting in blood pressure increase. Therefore, there is not much displacement and the acceleration is close to zero.
When the subject is breathing, the velocity signal contains distinct sinusoidal variation due to motion of the chest wall resulting from inhalation and exhalation. A polynomial cure fit was used to remove this low frequency noise. The integrated acceleration, which is the velocity signal in blue, has tenth order polynomial curve fit base subtraction.
Algorithmic Development
Detrending was employed to remove the lower frequency components of the signal. Specifically, a 10th order polynomial curve fit was incorporated after the first integration as a means to reestablish a flat baseline for the chest velocity signal. The velocity was then calculated. Even though the subject, in this case, holds his/her breathe the chest still moves slowly and can be seen to have substantial low frequency components. The corrected velocity is used to identify heartbeat time duration using the negative amplitude deviation segments. These negative peaks can be used to define a windows segment to be analyzed. Similarly, a correction using polynomial curve fitting is done to the displacement of the signal after the velocity is integrated. The polynomial function was derived as a least squares regression fitting to the velocity signal.
Three heartbeats were taken as the window of integration, using the velocity signal for the entire analyzed duration. The number of analyzed windows therefore equals the number of recorded heartbeats minus two. The interval of a window has a correction based on a ninth order polynomial curve fit and is performed on the velocity signal and on the displacement signal. Therefore, each heartbeat has a third order curve fit as a correction factor. The second heartbeat, which is in the middle of the window, is analyzed and provides the displacement of the chest wall.
The acceleration signal is integrated and provides the velocity signal.
In this case the analyzed data consists of seven seconds of recordings while the subject holds their breath. There are six heartbeats generating four displacement signals. The maximum displacement variation is associated with the chest volume change, which is directly related to the heart Stroke Volume (SV). The SV is related to the Ejection Time (ET) which is about 300 milliseconds from the first positive displacement peak to the second positive displacement peak (0.38-0.65 s). The Heart Rate (HR) is calculated using the time difference from one heart contraction to another. Therefore, the CO can be found by the multiplication of the average SV by HR.
The displacement signal at the upper chest wall varies much more than the displacement signal at the apex location and the displacement signal at the sternum location, and a peak velocity was not calculated by the program. Therefore, those values were chosen manually to calculate each displacement signal.
Even a very sharp filter cannot effectively remove the noise artifacts at low frequencies. Therefore, a polynomial curve fit based subtraction was employed to analyze a small window of three heartbeats, which is used after the first integration to provide the chest velocity. Another polynomial curve fit based subtraction was then applied at the second integration to provide the chest displacement. The average of the middle displacement of all windows provides reasonably reproducible information of the stroke volume and corresponding CO. In particular, the sternum location found to be better location to record the signal perhaps due to chest symmetry and the lack of fat and muscle at this area. Importantly, this location can be found more easily than the other two locations. Literature on the Kinetocardiogram indicates that the sternum location moves symmetrically inward which also justifies the sternum location [105]. The large and symmetric motion at this site has been explained by three factors: 1) the intrathoracic pressure change associated with ejection of blood; 2) a shift of blood from the lower to the upper chest; and 3) heart movement, pulling inward the anterior surface of the chest [105].
The mechanical activity of the heart is related to the electrical activity of the heart muscle, through a process referred to as excitation-contraction coupling. Correspondingly, the ECG can be used to provide a time marker to identify when the left ventricle is about to contract. To observe the heart's electrical activity, a pair of ECG electrodes may be used, which provide the second ECG vector, that is, in a direction from the right arm to the left leg. This vector direction reflects the hearts electrical activity from the sinus node to the apex, which is the natural pattern of the heart's electrical current flow. The heart's electrical activity shows six deflections. The P wave is associated with atrial contraction. The Q, R, S segment is associated with ventricular contraction. The T wave is associated with the heart's relaxation phase. The U wave is not common and associated with heart disease. Knowing these deflections, allows temporal alignment with the chest mechanical recordings and interpret identify the hearts activity.
Simultaneous recordings of the heart ECG and chest acceleration were taken to observe the relationship between the heart mechanical activity and its electrical activity while the subject was breathe holding. Three ECG electrodes were used to record the ECG lead two. The first electrode is placed on the right shoulder and the second is placed on the left side of the stomach below the left ribs. The third electrode is placed on the right side of the stomach below the ribs and is used as a reference potential. The Kistler accelerometer was placed on the sternum. Two filters were connected to the acceleration transducer. The first one was a high pass filter at 0.05 Hertz and the second a low pass filter set at 100 Hertz, to minimize noise and other recording artifacts. The accelerometer output was amplified by a gain of 100 before digitizing at a sampling rate of 2000 Hz. The recordings were taken with the subject in supine position. Duplicate recordings were taken. The first recording was taken with the subject's holding their breath, and the second was with normal breathing. It is known that the electrical activity of the heart occurs before the mechanical activity. Specifically, the QRS complex occurs immediately before the start of ventricular contraction, and correspondingly, the magnitude of the acceleration increases rapidly. Following the T wave of the ECG there is a period of high frequency vibration, which indicates the second heart sound and the beginning of the heart's relaxation phase.
The average displacement signal was computed and seen to be slightly different than the breath-holding displacement signal. In general, there is a greater chest displacement during cardiac contraction when the subject does not hold their breath.
Further analysis was done to correlate the displacement amplitude to breathing pattern. The velocity signal was used to distinguish the initial inhale and exhale periods. The breathing pattern is clearly seen in the velocity signal. However, it is hard to distinguish the pattern using the acceleration signal. This signal provides the inhale and exhale breathing periods, with the minimum velocity points identifying the beginning of the inhalation period. The maximum velocity points are considered to be at the beginning of the exhale period. These periods were identified and analyzed separately. The acceleration signal was integrated and filtered using high order digital high pass filter to reduce the breathing pattern. High order polynomial curve fit requires high computation power, and therefore it is not used in resource-constrained applications.
The average chest displacement due to heart contraction at the beginning of exhalation was found to be about 200 microns resulting in chest compression inward. During expiration, respiratory loading caused an increase in stroke volume. During exhalation, the intrathoracic pressure increases, resulting in decreased venous return, and therefore atrial filling, resulting in a decrease in stroke volume at the end of exhalation and beginning of inhalation; corresponding increase in heart rate. During inhalation, intrathoracic pressure decreases, enhancing venous return and therefore stroke volume, resulting in a decrease in heart rate decreasing during inhalation [127, 129, 130].
The average chest displacement signal measured at the beginning of inhalation in this sample is about 150 microns. As stroke volume normally increases during inhalation, this sample may be too early in the respiratory cycle to show the benefit of increased venous return.
Simultaneous recordings of ECG and acceleration signals provided a general interpretation of the displacement signal. Seismocardiograpy interpretation provided some information about the displacement signal but was not totally consistent with previous observations. Long duration recordings of chest acceleration allowed observation of cardiac differences between the inhale and exhale periods of breathing. The chest displacement signal was found to be different when subjects held their breath and when the subjects breathed regularly. In general, during an inhalation, chest displacement due to ventricular contraction is greater reflecting a greater heart stroke volume consistent with increased venous return associated with inhalation.
In the next stage of testing, a new, lower noise, and more sensitive, accelerometer was used to record the acceleration signal. The accelerometer 1221 from Silicon Design was used which provided greater sensitivity (2000 mV/g) and lower noise (5 μg/Hz1/2). The displacement signal, correspondingly, was observed to have a slightly different pattern than the previous recorded signal. In addition, wavelet analysis techniques were employed to remove the high and low frequencies components of the chest acceleration signal, and provide better artifact removal. The new transducer included amplification along with low pass and high pass filters. The high pass filter was set at 8 Hz and the low pass filter at 370 Hz. The recordings were performed similarly to the previous recordings. The displacement signal is slightly different, since the low frequency components below 8 Hz, which were captured in the previous recordings, have relatively large amplitudes. These low frequency components are reduced significantly with the 8 Hz high pass and so do not affect the signal as much as the previous recordings. In this case, all the signals are decimated to be the same length, and permitting better averaging. The average displacement is about 150 microns.
When a subject speaks, coughs, or vocalizes in any way, the chest vibrations overlap in the recorded frequency spectrum. Therefore, a low pass filter, as previously described, is applied at 50 Hz to minimize the influence of these artifacts. The significance of the filter is shown by looking on the Discrete Fourier Transform (DFT) of a typical acceleration signal before and after the filter. The sampling frequency at this point is 2 KHz, and is decimated by a factor of ten. In general, when speaking, women generate higher frequencies at a lower magnitude than men. This has an effect on the analyzed frequency spectrum. Therefore, the present example focuses on men. It is understood that an adaptive filter can assist in removing voice sounds and environmental vibrations and sounds from the spectrum to be analyzed. The frequencies over 50 Hz have higher magnitude during speaking. In men, frequencies in the 90-100 Hz range have high magnitude. The observed heart frequencies are primarily at 0.5-50 Hz. The frequency spectrum of a typical woman while speaking shows frequencies both lower and higher than 50 Hz have lower magnitudes than the observed in the frequency spectra of men. A twenty pole digital low pass filter at 50 Hz lowers the magnitude of frequencies associated with speech. The average displacement signal shows good correlation. However, not all the displacement signals align, but for the most part they do. Fast breathing or panting produces low frequency noises, and can be reduced by using high order high pass filter at 2 Hz.
The transducer provides three filters; the first filter is a three pole high pass filter at 8 Hertz; the second filter is a three pole low pass filter at 370 Hertz; and the last filter is a one pole high pass filter at 1.5 Hertz. The total gain of this system is about 400.
Because of the high pass filter, this transducer is only weakly sensitive to frequencies between 1 and 8 Hz. The observed cardiac frequency is generally considered to cover the 1-50 Hz range. Therefore, the input transducer may not contain all desired information. The transducer's transfer function is flat in range of 10 Hertz to 200 Hertz. The signal is sampled at frequency fs at 2000 Hertz, and decimated by a factor of ten, fd at 200 Hertz, before the wavelet analysis is done. Theoretically, the high pass filter should be at 2 Hz to reduce breathing and other low frequency noises, while the low pass filter should be at 50 Hz to reduce speaking and other high frequency noises.
Polynomial cure fit baselines subtraction demands substantial computational power. Therefore, more efficient and accurate methods were sought. Wavelet decomposition of the signal provides the capability to distinguish between different frequencies sets and reconstruct the filtered signal.
The analysis in this case utilized four steps. The first step was decimating the signal by a factor of ten. Therefore, the analyzed Nyquist frequency the original sampling frequency of 2000 Hertz became 100 Hertz. The second step was signal decomposition, where six levels of decomposition was performed using Matlab wavelet toolbox. It was found that the sixth level of decomposition using Shannon entropy as the cost function was most efficient. A wavelet program was written to have complete control on the processed signal and was used in the analysis. The third step was to reconstruct an output signal with selected packets base on the desired energy spectrum. The last step is to integrate the signal twice and acquire the displacement signal. Another analysis was done by preforming the double integration first and then performing wavelet analysis on the displacement signal.
Since the custom transducer had a high pass filter at 8 Hz, the velocity signal is centered at an equilibrium point after the first integration. The displacement signal, however, still has significant low frequency components. The acceleration signal is reconstructed from a fifth decomposition level using packets set of 2 to 28, where a sixth order Daubechies has been used as the mother wavelet. The double integration provides the reconstructed displacement signal, which is centered at zero. The low frequency component of the signal is reduced. Therefore, the ability of wavelet analysis to process the signal was found to be effective. Further analysis is needed to “polish” the displacement signal.
The displacement due to ventricular contraction is measured from the maximum peak around 0.28 seconds to the minimum peak around 0.32 seconds. All seven contractions contribute to the average chest displacement. The average does not represent the true average of all displacements. Therefore, the chest displacement of each heartbeat is calculated and averaged. Better results are achieved when the wavelet analysis is done on the true displacement signal, taking the double integration on the acceleration signal first and then computing the wavelet transform.
Clinical Testing
Ten subjects were selected for comparison recordings. The recordings were taken while the subjects were in the supine position and breathing normally for thirty seconds as well as during a short conversation of thirty seconds. This process was also completed while the subjects were in a seated position and a standing position. Wavelet analysis was performed on the displacement signal for each recording and compared. Most of the subjects were in their twenties, and two subjects were women. The average displacement of each subject is compared. Six decomposition levels are used and packets 2-28 were reconstructed to reduce the low and high frequency noises. Frequencies between 50 Hz and 100 Hz where not reconstructed in this example since there is no significant frequency content in the displacement signal in this range, and the literature also justifies using frequencies below fifty Hertz.
Since the raw signals are noisy, the algorithm uses the velocity signal as the marking point to find the ventricular contraction peaks. However, these peaks are not consistently detected.
There is an inverse relationship between chest displacement and BMI. As subject BMI is higher the chest displacement is lower. Assuming high BMI related to greater chest circumference, the total chest volume is greater; the chest displacement due to blood flow is smaller. As shows the R2 value is low, but the representative trend-line has percentage coefficient of variation of 26.6 from the base line. One subject was removed from this analysis since all other subjects were in their twenties. Better correlation is achieved when the chest volume and heart rate are factored into the regression analysis. The standard deviation percentage from the base line is better and the R2 value is higher at 0.46. Additional demographic parameters (age, gender, etc.) would need to be taken into account to provide an accurate estimate of cardiac output; nonetheless, this result shows the ability to use the measurements to assess cardiac activity. Moreover, some of the subjects needed to adjust the transducer on their chest to create more pressure between the transducer and their chest, adding errors to the recordings.
It is typically necessary to compare new measurement techniques to the existing technique to illustrate correlation between the two to provide proof of concept. If there exists a “gold standard” measurement, then comparison to the ‘gold standard” is essential. In the case of cardiac output assessment there is no existing gold standard. Invasive catheter based measurements are commonly used in the hospital setting as a central line has often been placed into a patient for some alternative purpose, but this approach is widely viewed as inaccurate, and moreover, preforming invasive cardiac output measurements is not possible in a non-hospital setting. Therefore, non-invasive cardiac output monitoring (NICOM) equipment was used to provide cardiac output measurements. It is important to perform simultaneous recordings while comparing the measurements to show a “standard” measurement. Specifically, a bioelectroimpedance based technique developed by Cheetah Medical of Israel was elected, which had recently received FDA approval.
Largely as a result of NASA funded research, bioelectrical impedance techniques for estimating cardiac output have been shown to be an effective alternative to ultrasonic or invasive measurement approaches to obtain CO. Correspondingly, over the last decade, several companies have begun to offer commercial CO monitoring devices based on bioelectroimpedance. Specifically, Cheetah Medical has developed bio-impedance system (NICOM) which they refer to as a bio-reactance device. This device has obtained some acceptance in the hospital environment and provides continuous non-invasive cardiac output monitoring for several hours, or until the electrodes become detached from the skin. Importantly, this equipment does not require a physician or other clinician as an operator, significantly lowering operating costs. Standard bio-impedance systems rely on a standard four-electrode current source recording arrangement. They apply a high-frequency constant amplitude electrical current across the thorax using two electrodes, and record the corresponding voltage difference between the remaining two high input impedance electrodes. The ratio between the measured voltage and applied current amplitudes is a measure of transthoracic impedance. This instantaneous impedance change is related to the stoke volume (SV) change, and is proportional to the product of peak flow and ventricle ejection time (VET). SV is proportional to the product of maximum impedance change and to the phase shift change.
The correlation between chest motion due to heart contraction to cardiac output was determined. The first verification was performed between the reconstructed infrasonic displacement signal and the NICOM measurements of cardiac output. NICOM, ECG and Infrasonic measurements were taken simultaneously while the subject was in supine position; seated at an angle of 30°, supine with legs are at 30° from the horizontal line parallel to the ground, and after a short exercise. The average of the sternum displacement, has been compared to NICOM cardiac output measurements. The sternum displacement was measured at the two hundred milliseconds time point following the ECG QRS complex, calculated from the reconstructed infrasonic displacement signal. In this case, a 10s average sternum displacement obtained from a one minute NICOM cardiac output recording is compared.
The chest acceleration signal is recorded simultaneously with the ECG signal and is integrated twice. Wavelet transform is performed on the displacement signal, where the original signal is decomposed and specific packets are selected for reconstruction. The reconstructed signal is then aligned with the ECG signal. The inward movement of the chest is captured by the reconstructed displacement signal.
Correlation analysis was performed using the average chest volume change due to heart contraction over one minute interval. The cardiac output is the product of the average volume change per minute and heart rate obtained from the Infrasonic cardiac output.
A correlation of R2=0.72 was achieved by computing five wavelet decomposition levels using 6th order Daubechies as the filter coefficients. At the fifth decomposition level, packets two to twenty eight were selected for reconstruction. Shannon entropy indicated that the most signal information is within the first packet, which contains the lowest frequency components, but these are below the heart's frequency spectrum. Therefore, Shannon entropy does not provide a good indication for selecting packets for reconstruction. Packets with the most heart information are between one to fifty hertz and so were selected for reconstruction. Another aspect for reducing computational time and increasing cardiac output correlation, is by computing more decomposition levels. By computing more decomposition levels finer frequency segments and select a better packet set is achieved. However, computational time increases as decomposition level increases. The R2 value does not change much between fifth and eighth decomposition levels, where all the packets were selected for reconstruction except the first packet; lower frequency range, and the last four packets, high frequency range.
Choosing a different Mother wavelet may also increase cardiac output correlation and decrease computation time for lower order filters. Each filter has double its coefficient based on its order. For example; Daubechies one, which also known as the Haar wavelet, has two filter coefficients; Daubechies two has four filter coefficients; Daubechies three has six filter coefficient and so on. From Daubechies two to Daubechies ten the correlation to NICOM cardiac output is about the same. Daubechies two is computed the fastest, but in general the size of the filter does not greatly affect the computation time nor provide better results.
There are many possibilities to perform different filters at different decomposition levels and choose different packet sets to achieve faster computation and obtain better cardiac output correlation. The possibility of using one or more packets to correlate to cardiac output is also an option which should be investigated. Since there are so many potential wavelet packet possibilities, Genetic Algorithm strategies were applied to define the best packet set or CO prediction.
Since one measurement does not provide a reliable cardiac output correlation, further investigation was done on multiple subjects. Measurements from four subjects were taken simultaneously using the NICOM, ECG, and chest acceleration. Subjects were asked to be in supine position for seven minutes until NICOM calibrated and performed sufficient CO measurements to start the experiment. Subjects were moved to a sitting position, and then began to exercise for 35 minutes, involving cycling for two minutes and resting for five minutes.
Previously the ECG algorithm was able to detect the contraction time (QRS complex) when the ECG signal was stable, but did a poor job when signal was not stable; i.e. when the subject was exercising. The algorithm was modified to have better QRS detection using a wavelet transform. Similarly to the acceleration signal, the ECG signal suffers from low and high frequency noises. A similar approach was used to remove the noise from the ECG signal, providing a correlation of R2=0.99. Good CO correlation of R2=0.84 was found, but a closer look at SV and HR indicates deficiency in SV correlation.
The transducer captured the acceleration signal from the sternum. This signal is filtered using a low pass filter at 50 Hz to remove high frequency noises, since the heart motions are largely below this frequency. The signal was decimated by a factor of ten originally, and during NICOM measurements was decimated by a factor of twenty. The acceleration signal can be converted to a displacement signal before or after the wavelet analysis. When NICOM measurements were taken, the acceleration signal was converted to displacement before computing the wavelet analysis. The number of decomposition levels and the wavelet filters are set before computing the decomposition. The packet selection is done on the last decomposition level. The displacements of the heart contractions are captured from the reconstructed displacement signal and averaged over a period of one minute. Cardiac output is calculated and compared to NICOM recordings. Use of a Genetic Algorithm may provide better correlation to NICOM by selecting different packet set.
The initial recordings utilized an off the shelf Kistler accelerometer to measure chest displacement during heart contraction. The literature on kineto-cardiograms justified the sternum recording location by showing symmetrical inward motion [105]. Moreover, this location is justified due to its physical structure; the accelerometer can be placed on the sternum where there is little muscle or fat and is found easily. The recordings at this location were found to be consistent when using a polynomial curve fit baseline subtraction and integrating the result to find the displacement of the sternum. This analysis was performed on both the velocity signal after integrating the acceleration signal and the displacement signal after integrating the velocity signal.
Simultaneous ECG recordings confirmed the initial proposal that the main negative deviation of the displacement signal is due to ventricular contraction. The simultaneous recordings also allowed us to identify the infrasonic cardiac output deviation and distinguish inhaling cardiac function from exhaling cardiac function. Chest displacement during inhaling is found to be greater than inhaling and is consisted with literature [129, 130]. Polynomial curve fit base subtraction provided good filtering tool, but requires significant computation power. Therefore, wavelet transform analysis is more suitable.
A custom made transducer was developed to improve chest infrasonic acceleration recordings due to heart contraction. This transducer provided lower noise measurements with higher sensitivity. Wavelet transformation was able to eliminate noises from the recorded signal. The analysis was performed on ten heart beats on ten subjects and observed that the variance on the ten displacement signals was about 30%, but better correlation was achieved when BMI and other physical differences were incorporated into the analysis; specifically, a 25% variance was obtained when factoring in Heart Rate, Chest volume and BMI. Better results can be obtained when more components are factored in and better signal analysis is performed.
Finally, a comparison between the infrasonic measurements and an approved non-invasive Cardiac Output monitoring were undertaken. Comparison to the NICOM demonstrated good correlation, R2 value of 0.72.
In the past, wavelet transform and GAs where combined yield results for the problem set they were used. In this case, non-traditional wavelet computation is employed, where just decomposition is performed and a GA is used to define a specific packet set which correlated best to the ground truth. An initial method did not work and further investigation was done to modify the algorithm to identify a desirable solution. A series of experiments was used to test the algorithm, and after restricting the correlation value R2, the algorithm was able to converge. The final algorithm was used to identify specific features that correlate best to NICOM SV giving four subjects data.
In this application, a subcomponent of chest wall motion (seismocardiogram recording) is sought to be discovered which can be used to estimate a specific activity of the cardiac muscle, for example, stroke volume. The time consuming operation of waveform reconstruction is sought to be avoided, since the application calls for rapid response from a resource limited device. Moreover, there is a potential to investigate for better correlation.
SV is estimated from chest acceleration, at the xiphoid process [108, 111]. The approach involves performing multi-wavelet decompositions on the acceleration data to generate a large pool of features from which the GA is used to select the best packet combination for predicting SV. The “ground truth” SV is obtained using electrical impedance based Cardiac Output Monitoring device NICOM.
Eshelman's CHC GA [147] search engine combined with the MMX crossover operator [148] identifies the best subset genes (i.e. packets), from a multiple filter bank. Since the goal was to minimize the number of genes to avoid over fitting and to reduce the computational costs of SV estimation, a Sub-Set-Size (SSS) variable was defined [149] and added to the chromosome.
Only when it becomes clear that further crossovers are unlikely to advance the search, a soft restart is performed, using mutation to introduce substantial new diversity, but also retaining the best individual chromosome in the population.
The initial GA population is generated randomly using a uniform distribution. In CHC two initial populations are produced and the chromosomes are evaluated, and the more fit chromosomes from both populations are selected to become the next population. For all subsequent generations, the pairs of parents (randomly mated) produce two offspring and the selection operator produces the next parent generation by taking the best from the combined parents and offspring using simple deterministic ranking.
Understanding the chromosome structure provides an understanding of the connection between the feature-genes and the Sub-Set-Size (SSS) gene. A chromosome is defined as set of genes, and in this approach, the first gene represents the SSS, that is, the number of genes that are expressed when a chromosome is evaluated (
The expressed genes in a chromosome represent the magnitudes of a subset of wavelet packets. The mathematics of the wavelet transform may be found elsewhere [125, 126, 127]; here discreet wavelet transforms (DWT) are used. In wavelet transform analysis, the focus is often the low frequency components. The time sequence is separated into two components: low frequency components, called approximations, and high frequency components, called details. Subsequent levels of decomposition are performed on the approximation coefficients; again separating the low frequency components in to approximations and details. This process is repeated with entropy, energy, and/or a cost function being computed after each level of decomposition as a means of optimizing the decomposition process.
In cardiac analysis, the acceleration data may include numerous high and low frequencies not associated with cardiac activity. High energy at the low frequency is likely to be associated with breathing and whole body motion, while high frequency components may be associated with vocalization. Since the goal is to identify those components providing the best correlation with SV, the full signal frequency spectrum was investigated regardless of its computation cost, energy, or entropy.
Full tree decompositions, that is decomposition was performed on the details and approximation coefficients of each branch using one Mother wavelet (
An ECG signal was used to capture the ventricles contraction time (QRS complex), which serve to identify the time point to evaluate in the decomposed acceleration signal. Four decomposition levels were performed with six different mother wavelets providing ninety six different features associated with ventricle contraction acceleration energy.
The goal of utilizing the subset selection GA was to identify the minimal subset of features capable of accurately estimating the NICOM reported SVs. The NICOM provides thirty-second averages of SV and so wavelet decomposition was performed on each thirty seconds of recoded acceleration data. Eighty-five thirty-second averaged measurements were taken sequentially using the NICOM, the ECG, and chest accelerations, from a single subject during both rest and during exercising. There were five exercise periods for one hundred and fifty seconds at the same intensity and five resting periods of two hundred and seventy seconds. Data was collected while subject was at rest, in upright position for four hundred and fifty seconds. Multivariate regression was used to correlate the expressed chromosome genes ‘packets energy’ to the averaged NICOM SV measurements. The R2 value of the regression line was used as the chromosome fitness value. The higher the R2 value, the better the gene set predicts the NICOM SV.
In the CHC GA, the more fit chromosomes remain in the population until they are replaced by even more fit offspring. The fitness functions returns a two-vector, where one is the R2 value, and the other is the SSS. The SSS is located at first chromosome gene. The vector selection process works by comparing two chromosomes, a parent, A and an offspring B, if R2(A)>R2(B), then A is more fit (and vice versa). However, if R2(A)=R2(B), then the chromosome with the smaller SSS is more fit. If the SSS's are also equal, the parent is not replaced.
The crossover operator is responsible for offspring reproduction. It consists of three operators: Incest Prevention that decides if the two parents can mate; Index Gene Crossover that is responsible for inheritance of both parents' genes to the offspring; SSS Recombination crossover that is responsible for setting the SSS gene of the offspring based on both parents' SSS genes.
The crossover operator is applied to each random pair of parents. The first step is to check the pair for incest prevention. Parents who are too closely related are prevented from mating. The distance between two chromosomes is simply the number of unique genes, in the leading portion of the chromosomes out to the furthest genes an offspring might inherit (the larger value of SSS genes from the two chromosomes). The initial value for the incest threshold is half of the maximum SSS, but it is decremented whenever a generation occurs in which no offspring survive. When the incest threshold drops to zero, any chromosome may mate with any other, including a clone of itself. The incest threshold dropping to zero is one of the criteria used by CHC for halt and restart decisions. This incest prevention algorithm has been shown to effectively defeat genetic drift [168]. It does this by promoting exploration, allowing only mating among the more divergent chromosomes; as long as this process is successful (offspring survive). Being self-adjusting, it tunes itself to problems of differing difficulties; when more fit offspring are being produced, the threshold remains fixed, it drops only when progress is not occurring.
GA research has shown that “respect” is an important property for a crossover operator [199, 200]. That is, if the parents share common genes, it is important that the offspring should inherit them. The MMX_SSS operator achieves this by first copying the common genes from the parents to the offspring. However, given that there is selection pressure for smaller SSS gene values, this copy operation moves each gene one position forward, to the left, in the offspring (
The last step in crossover is to set the values for the SSS genes in the offspring. This operation uses the “blend crossover” or BLX [149, 198]. The SSS gene for each offspring is drawn uniformly randomly from an interval defined by the SSS genes in the parents and their fitness (
The common genes from the two parents are copied one space to the left in the offspring and the other genes are randomly inserted into the offspring. In this example, the first parent common gene 51 switches places first with gene 12 and then gene 87 in the next generation, (offspring one) because all three are common in both parents. Gene 69 from the second parent stays in the first place since gene 41 is not common (offspring two). The rest of the genes, the “unique” genes, are copied to a grab bag, the table on the right in
The interval is first set to that bounded by the parental values, and then extended by fifty percent in the direction of the more fit parent. In the example illustrated in
To evaluate this approach, a series of experiments were performed to test each aspect of the algorithm; these experiments are described in sequential order. All experiments used seismocardiogram data from a single subject obtained at rest and while undergoing mild exercise (light bike pedaling in an upright position with back support). Four levels of wavelet decomposition were performed on successive thirty-second time intervals. Six mother wavelets were utilized: Daubechies, Symlets, discrete Meyer, Coiflet, Biorthogonal, and reverse Biorthogonal. A “ground truth” SV value was obtained for each thirty-second interval from the NICOM. This produced a data set with 96 features (6×24), and a “true” SV for each of the 85 intervals that were measured. The maximum value of SSS was set to 32 assuming the GA could obtain results with a subset much smaller than this. Thus, the chromosome contained 33 genes, one for SSS and 32 packet indexes. For fitness to maximize, the R2 from a linear regression of the packets energy to SV was selected. The population size was one hundred, the number of soft restarts was set to ten, with maximum zero accepts (restart condition) set to three.
The first experiment was directed toward achieving a maximum R2 value, but showed little evidence of convergence.
Failure of convergence from experiment suggested verification of the algorithm. A perfect solution was embedded in the data, to test the algorithm's ability to discover it. A set of five features was selected and their values “doctored” so that together they have perfect SV correlation. These features had indexes of 4, 31, 67, 80, and 92. (i.e., widely distributed among the pool of features). The “doctored” features emerging as the only genes left in the population after about one hundred generations (
The algorithm was then challenged by perturbing the data with Gaussian noise, where each feature is the original value plus twenty percent Gaussian noise. The characteristic pattern of convergence failure was observed (
To test the influence of R2 on convergence, the number of significant digits in the value of R2 reported by the regression to the GA was reduced. By setting this to two significant figures, it was declared that chromosomes that differ in R2 by less than 0.01 should be considered equivalent, thereby allowing for ties and enabling the second level of the hierarchical fitness selection to kick in. One may also think of this as an admission that an R2 estimated from a sample of cases must of necessity contain a certain amount of noise (sampling noise rather than measurement noise); allowing the GA to over-exploit noise provides no benefit. This strategy resulted in a return of effective performance even though the problem is now more difficult because of the noise perturbation (
Having an indication that over-precision was precluding convergence in the presence of noise, the original dataset was rerun with R2 reduced to two significant digits. The patterns that indicate successful learning was observed, and this time without the presence of doctored data. Now SSS evolves, first to 22 packets (in the first convergence, and the next eight soft restarts) and finally to 21 and 22 in the last two soft restarts (
The CHC genetic algorithm with the MMX_SSS crossover operator has previously been applied to the task of feature selection in bioinformatics classification tasks. This algorithm may also be applicable to feature subset selection tasks in time series data processing, but the use of a high-precision first fitness metric such as R2, seems to require a judicious reduction in significant digits provided to the GA in order to induce ties so that the second metric (SSS) may become active. In classification tasks, ties are common since counts of classification errors have a limited dynamic range. This shows that a tradeoff between sensitivity to small improvements in accuracy and the desire for small subsets is appropriate.
This algorithm can be applied to selecting high performance, small set of signal features that can be combined to yield accurate metrics of some signal content. Finding specific mother wavelet packets that can be combined at the energy level without full waveform reconstruction can enable computationally inexpensive ways to extract information from time series data.
The CHC genetic algorithm with the MMX_SSS cross-over operator has previously been applied to the task of feature selection in bioinformatics classification tasks. Evidence is provided that this algorithm may also be applicable to feature subset selection tasks in time series data processing, but the use of a high-precision first fitness metric such as R2, seems to require a judicious reduction in significant digits provided to the GA in order to induce ties so that the second metric (SSS) may be-come active. In classification tasks, ties are common since counts of classification errors have a limited dynamic range. This work seems to show that a tradeoff may be needed between sensitivity to small improvements in accuracy and the desire for small subsets.
The last experiment yielded good correlation and as results the same algorithm and settings are used in this case to find a solution for four subjects. The filter bank was expanded to 640 features derived from different mother wavelets and another six features derived from subject physical measurements (Chest volume, Chest circumference, height, weight, BMI, BSA). The GA population size was increased to 200, allowing farther exploration of the landscape for the optimal solution. Similar to the previous experiment
Finding the contraction time location using the acceleration signal is challenging compared to extraction from an ECG signal. As described above, the ECG R-wave was used to define the contraction time location to extract values from the filter set using a regression line to compute the SV. However, cardiac parameters including the contraction time, can also be estimated using only an accelerometer. GAs are also used to find a global solution. A computationally efficient method is provided.
Extracting the timing of heart contraction from acceleration data at the chest wall using a standardized algorithm for all subjects is challenging, because the chest acceleration signal is individual based on body characteristics, since each individual chest vibrates differently when the heart contracts. Moreover, the chest vibrations due to the heart contraction are affected by breathing motions, speech and other motions. The subject heart acceleration may also vary from one heart beat to another. The ECG R wave is clearly distinguished in all subjects where the first heart sound within the acceleration signal of each subject varies in amplitude.
The low pass filter was set to 50 Hertz and the high pass filter was set to 2 Hertz.
The ECG QRS complex function was used to extract packet information at the heart contraction time location, as the ground truth for heart contraction time location. Based on physiological assumptions, a time segment was chosen after the ECG contraction time location to serve as the window of opportunity for capturing the heart contraction time location via acceleration data. True Positive (TP) detection is considered as heart contraction detection based on the acceleration data in this window. Otherwise, if no heart contraction is detected a False Negative (FN) accrues. If a heart contraction is detected outside of the window, the detection is considered as False Positive (FP).
The heart mechanical activity follows the electrical activity. The time lag of the heart contraction and accelerometer electrical circuit after the ECG R-wave is about 50 milliseconds. The effective time lag is dependent on filter delay. Since the data was analyzed at 100 Hertz and four decomposition levels performed, the total time scaled energy observation of a packet per data point is 160 milliseconds. Therefore, the window in which a TP has occurred is equivalent to the same time scale window following the ECG R-wave. Optimal detection means that all heart contraction time locations from the acceleration signal are located in the TP windows, and there is no heart contraction detection elsewhere. The Sensitivity and the Positive predictive value were measured and calculated.
Two approaches were investigated to detect heart contraction from the acceleration signal; the Discreet Wavelet Transform (DWT), and the Continuous Wavelet Transform (CWT). The CWT calls for more redundancy and may provide more features which allows easier detection. The DWT calls for better noise elimination, where signal components can be eliminated. Both approaches use a detection function and evaluation function which compare the detected contraction time location to the ECG QRS time stamps. In general, the DWT convolves the input signal with specific filter coefficients and decimates the signal by half to eliminate redundancy. After one level of decomposition the approximated signal is half of the input signal length. The next decomposition will convolve the approximated signal against the same mother wavelet low pass filter coefficients. The second decomposition level investigates a narrower band of low frequencies than the first decomposition. Most likely, the best frequency detection occurs when the mother wavelet filter coefficients represent the input signal. In this case, that happens at higher a decomposition level, when the mother wavelet is similar in shape to the input signal.
The second approach is to use the CWT to capture information which the DWT may miss. The CWT calls for redundancy since the frequency component of the signal is redundant after performing convolution. The CWT uses the actual mother wavelet coefficients as opposed to the DWT that uses the Multi-Resolution Approximation (MRA) equation. As the scaling function increases the number of the mother wavelet coefficients increases. In general, to capture the desired signal information the mother wavelet shape should match the desired information shape (i.e. similar frequencies) and this is done by choosing the correct scaling.
Both CWT and DWT are good filtering tools. They are similar approaches, but have different advantages and disadvantages. Since, with the DWT multiple decomposition levels are available, the option of sharper filtering to capture specific frequency components is possible, noise is better reduced than with the CWT. The CWT does not compress the input signal for sharper filtering. Instead the mother wavelet “stretches”, requiring more computations than the DWT, and so redundancy of the frequency components may provide better feature detection.
The output data from the DWT brute force and CWT brute force functions were processed via a detection function that detects the heart contraction time location. The contraction detection function output is then evaluated using the evaluation function discussed above, based on the ECG QRS heart contraction time location. In one embodiment, a function divided the processed signal to many segments or “windows”. Each window was evaluated by its maximum energy peak which was compared to an average threshold number. The average threshold was set to half of the averaged last ten peaks. After the CWT threshold algorithm was tuned, each threshold window then consisted of the positive segments of the CWT threshold output.
The DWT brute force algorithm evaluates all the possibilities to compute a solution for heart contraction time stamp. Each Mother wavelet packet combination set is evaluated. In this case, there are two loops. The first one changes the mother wavelet selection and the second changes the packets combination selection. In each evaluation the input signal is decomposed to a three decomposition level, packets are selected for reconstruction, and the reconstructed signal is processed by a computational function before it is evaluated by the peak detection and evaluation functions. Each choice of MW and packet combination was stored in a chromosome structure. The selected MW is at the first position, the computational function is at the second position, and the last eight positions are occupied by packet reconstruction selection. Each MW was assigned a number which was parsed using a parsing function. The Computational Function (CF) was set prior to the run and was applied on each MW packet combination. The packet combination used eight characters of ones and zeroes to define the selected packets for reconstruction; one selects the packet and zero ignored the packet.
The maximum detection from the brute force run was about 97 percent, but the second heart sound was also sometimes counted as a contraction location. Therefore, the detection false positive rate was about 50 percent and the calculated heart rate was double the measured heart rate.
Good results were obtained using two subjects. However, when the analysis was run on a third subject it failed to detect the heart contraction time, because the threshold function eliminated the heart construction segment. Similar frequencies where associated with the third subject heart construction, first heart sound, and the second heart sound. Therefore, it was concluded that a Genetic Algorithm was needed to generalize an algorithm to fit all subjects.
A Genetic Algorithm can be a useful tool to discover the global optima solution or a solution which is close to it in a large landscape. Since there are deviations among subjects, a large population of subjects is required to formulate a generalized algorithm, e.g., four females and eight males. Basic information was collected from each subject (like height, weight, age, and etc.), followed by collecting acceleration data for ten minutes while the subject was in a supine position, ten minutes while subject was in an upright position, four minutes while subject was in an upright position and talking. In the following GA detections, DWT and CWT, only three minutes of the male subjects' data was analyzed while in an upright position. Computing multiple filters and evaluating the solution was very computationally expensive. Therefore, selecting a portion of data from each subject that is sufficient to represent deviation within the full data spectrum (all subjects) is expedient. Three minutes were selected as a sampling duration, to record at least a hundred heart beats for each subject to have confidence in the solution. Since data of one subject were not collected correctly, it was discarded. A total of seven male subjects in an upright position and three minutes of recording while sitting quietly were analyzed where the ECG signal was clean of noise and the R wave was fully detected.
A chromosome structure was provided to evaluate the selected packets, MW, Computation Function (CF), and threshold function. Previously, the CF and threshold function were set to be constant. Here, the GA chooses the best threshold function and CF to maximize detection. Also, multiple wavelet transforms were combined to provide better detection by the GA. The CHC GA was used again because of its robustness. Here, the crossover operator and chromosome structure were modified. An example chromosome provides two MWs at two decomposition levels. MWa has a threshold function THa and corresponding packets aB0-aB3. MWb has a threshold function THb and corresponding packets bB0-bB3. The CF computes the output combination for the two.
The computation function combines and performs mathematical operations on each wavelet transform. Here, five functions were available to each chromosome. Those functions were chosen based on some assumptions and for being different from each other. Therefore, a function that provides a good result within the Evaluation Function will rise quickly and eliminate others. Each function does element operation on MW Signals (MWS) after decomposition, thresholding, and reconstruction.
The threshold function performs mathematical computations on the decomposed wavelet transform before reconstruction. The purpose of this function is to eliminate noise and focus on the features that are associated with the heart contraction. The packets to be reconstructed are equal to a function of the decomposed packets. In some cases, a threshold value is set at the beginning of the run to save searching time.
The CHC GA was used again to converge to a near-global optimum solution. However, a different crossover was used to generate offspring since chromosomes had a different structure. In the reproduction process a bit representation was used for the packet selection and a numerical representation for the MW, threshold, and CF representation. The HUX (Half Uniform crossover) was used for packet selection crossover, since it has general been observed to perform well when using bit-wise operations. The common genes transfer to the offspring and their location do not change oppose to SSS_MMX crossover, where the common genes moves one step towards the beginning of the chromosome. The rest of the genes are processed as follows, where half of the unique genes are chosen randomly, line under, to change their state (switch to the opposed binary state.)
The blend crossover (BLX) crossover was used on the MW, threshold, and CF genes. This crossover was used before with the SSS_MMX crossover on the subset size gene. The BLX crossover formula is given below where Gene Parent one (GP1) is smaller than Gene Parent two (GP2).
If the upper bound of the Interval is greater than the number of features, it is set to be equal to the number of features. If the Interval lower bound is smaller than one, it set to one. Gene Parent one (GP1) and Gene Parent Two (GP2) represent the rage which is extended by 50 percent to the direction of the more fit parent GP2, where a random gene can be selected.
The DWT GA evaluation function was similar to the brute force DWT and it included many function in it. The evaluation function reads the chromosome structure and sets the packet selection parameters using the parse function from the brute force DWT evaluation to define the MW function. DWT decomposition is computed, then the threshold function is computed on the selected packets, and the waveform is reconstructed. The cost to compute the DWT is calculated using the cost function. The computation function (CF) does element mathematical computation which provides better detection. The output from the CF is then multiplied by an initial CWT threshold function which determines a vague window of the first heart sound time segment. This window is used to eliminate the second heart sound. The processed waveform is then transferred to a peak detection function which determines the contraction time location. Those time locations are compared to the ground truth ECG signal and TP, FP, FN are calculated. A sigmoid evaluation function was computed which also has been used to evaluate each chromosome. Better results were found using the original evaluation function.
The main purpose of this algorithm is to isolate the first major acceleration deviation from the second one (in phonocardiogram, first heart sound S1 from the second heart sound S2). As a result, this algorithm is able also to detect subject heart rate, number of beats per minute. The chest acceleration signal is more challenging than the ECG for heartbeat detection, since each individual chest vibrates differently when the heart contracts. Moreover, the chest vibrations due to the heart contraction are affected by breathing motions, speech and other motions. This function is used after the brute force approaches had difficulty in isolating the heart contraction phase from the relaxation phase due to heart valves closing sounds.
The CWT threshold function defines the segments to be analyzed. It starts by initializing the HR to 60 beats per minute. The CWT scaling is calculated based on the HR and different scaling is selected based on the HR.
The selected scale is used to scale Daubechies five MW which selects a window where contraction occurs. Originally, the DWT GA detected twice as many heart contractions as were measured. The Evaluation function was modified many times to achieve better results but the DWT GA was not able to provide a good solution since HR is so different from one subject to another. The positive predictive values were around fifty percent. Since this function depended on the HR, it solved this issue. Heart rate is measured by counting the number of threshold windows per minute, and is used is scale the CWT function accordingly. The sigmoid evaluation function uses the ECG signal to generate a sigmoid like function around a small window after the heart contraction occurred.
A heart contraction time location at this window will result in no penalty. If the detection occurred at the edge of the window, a small penalty is added. If detection occurred outside of the window the full penalty is added. If no detection occurred, a no-detection penalty is added, which is greater than a bad detection penalty. The window consists of both the sigmoid equation and its flipped version where the variable X starts from −2 to HWS, (Half Window Size), for smoothing purposes, using the equation above. After the whole signal is analyzed, the penalties are added. The GA is minimizing the sigmoid function, where the most fit chromosome has the smallest penalty.
The solution which is provided here resulted from many iteration and modification of the DWT GA, evaluation function, chromosome structure, and more. On average a full run takes several days. Note that this optimization is not performed at the time of use in the target system, and therefore time for optimization is not a limiting factor.
In this case, the search was restricted to receive an answer in a week. The number of chromosomes in a population was set to fifty and the number of MWs functions was set to one at five levels of decompositions. Data of three minutes from each of the seven subjects was used to correlate the heart contraction time location. The DWT parse function was used again and was modified to a smaller number of MWs.
After the second soft restart, the GA was not able to converge to a final solution until it hit the maximum number of generations. The GA was able to identify a MW function that best suits the converged packet combination, evaluation type, and threshold function type. This run included four evaluation functions and seven threshold functions.
Note that, for any given subject, if calibration data is available, such as CO from a NICOM unit, then the algorithm may be tuned to that specific person.
The evaluation function is intended to maximize the detection; therefore, more weight was given to the detection of a heart contraction time location than to wrong time location detection. During the run the sigmoid function and the evaluation function were used. The evaluation function was modified to specifically weight sensitivity and positive predictive values.
The evaluation of the best chromosome from this run was detection (sensitivity) of 98.62%, and positive predictive value of 98.56%. That means that the detection was mostly at the right time and at the right location. This solution is sufficient to determine CO and average SV. Note that the “gold standard”, thermo-dilution has ˜80% accuracy and NICOM has ˜65% correlation. Missing a heartbeat in a minute is at most 2% from 100% detection. Therefore, the provided solution is useful and sufficient for most purposes.
One of the key components is to eliminate the low frequencies from the collected data since they provide an offset noise. This solution eliminates the lowest packet which includes those frequencies, and provides a satisfactory solution. Moreover, this solution eliminates more than half of the packets and those packets are next to each other, reducing computational cost.
A final goal was to create a prototype that uses a low power microcontroller. The less computation required, the less power is required, the smaller the microcontroller can be, and longer monitoring is available. Therefore, the optimized solution provides a good solution to detect the heart time contraction.
A second method to determine heart contraction time location is the Continuous Wavelet Transform (CWT). The CHC GA was used to determine the best filter set to extract the heart construction time location. Similar to the DWT GA, a computation function was used. However, the computation function was set to be constant and was changed manually, from one run to another. Two different types of Gas were performed. In the first GA, the evaluation function was a regression line based on the chromosome genes, and the second GA was a convolution-based approach for each of the chromosome genes. In the first CWT GA, the features were a result of the CWT output using multiple scaling and MWs. A GA as discussed above and its crossover were used to determine the best filter set that provided the optimal heart contraction time location from the acceleration data based on the ground truth, the ECG data. The same evaluation function as the DWT GA was used to evaluate each chromosome.
This GA was run on a server farm with 24 cores, to speed up the GA process. The run took three days to converge to a solution with multiple soft restarts. Matlab, CWT function was used to compute the data base (features) before the GA process. Then, the CWT GA searches for the optimal solution within the data base. Each of the CFs used in brute force approach where used.
This GA offered two appealing solutions. The first solution has better sensitivity where the second has better positive predictive value. The first solution used fewer filters than the second solution. The first solution had a sensitivity of 0.9862 and positive predictive values of 0.9869 and performed element multiplication of the selected filters. The solution consisted of two filters which is a reasonable solution for a microcontroller with limited computational power. In both cases, the maximum SSS was set to sixteen genes, which provided a good search base, where the GA was able to converge to a smaller SSS. The first solution was able to converge on the optimal solution five times.
In some cases, the GA is not able to converge. In this case, on the fifth soft restart the GA was not able to converge which resulted in reaching to the maximum number of generations (10,000).
This first solution provides two features (filters) which together provide the optimal solution. Those two features were sampled more often than the rest of the features, which indicates strong connections between them.
The second solution consisted of three filters and provided a better positive predictive value than the first solution. However, it has lower sensitivity. Here, summation of each of the filtered signals was performed, which used more filters by doubling the scaling of each MW. Also, in this solution, the filters (features) which contribute to the optimal solution were sampled frequently, but not all were sampled the most. The features with the strongest connection rose first, but the features that contribute to the global solution, which took generations to evolve, were not necessarily sampled more often. In this case, feature 686 was sampled more often than feature 6191 which was used in the global optima solution.
The two solutions provided satisfactory results, where sensitivity and positive predictive values were highly correlated to the ground truth ECG contraction time location. Both solutions used few features enabling the required computations on a small microcontroller.
In some cases, it is common to convolve two or more filters to observe specific frequencies and eliminate noise. Therefore, the same method was employed here, and a Convolution GA created. This GA convolves all the filters within a chromosome based on the subset size gene, allowing multiple convolutions to be performed and evaluated, using the same CWT filters. Chromosomes with many filters do not survive due to the convolution outcome.
The initial brute force approach did not result in a satisfactory solution. The DWT approach was not able to determine a good solution, but was not run with a large number of features due to computation time. The CWT approach provided a good solution for two subjects, but was not able to generalize the solution to more subjects. Also, like the DWT approach small numbers of features were tested due to computation time. It is important to notice that the run of each approach took more than a day to compute. Therefore, using a GA to search for a solution in a much larger landscape seems to be the right approach to continue.
An Advanced RISC Machines (ARM) microcontroller operating using the mbed.org environment were selected for fast prototyping and performance.
The sensitive and low noise Silicon Design Model 1221 accelerometer was used throughout the early experiments, which allowed accurate recordings and identify desired features within the acceleration signal. The signal was processed using the Bio-pack M-150 data acquisition system, which has 24-bit precision analog to digital converter. It was found that lower precision would suffice, and therefore a 16 bit ADC could be used. At least a 32 bit word and 16 bit precision should be used in the calculations.
A microcontroller consists of a microprocessor, memory, clock oscillator, and input and output capabilities. Therefore, it is possible to use it without extra components comparing. As opposed to ASIC, MCUs are not customable, and have functionality limitations. MCUs perform only digital computations, and so an Analog to Digital Converter (ADC) is necessary as an input device to read analog signals. MCUs are out of the box working solutions which are provided with datasheet, drivers and code examples. They are good in implementing difficult algorithms. Their main advantage is low upfront cost, ease of programming (usually programmed in C/C++), and relative low power consumption. In the past few years ARM (Advanced RISC (reduced instruction set computer) Machines), has acquired big portion of the MCU market. This technology is wildly used in embedded devices such as smart phones, which may include Bluetooth, WI-FI, LCD or OLED display, variety of physical sensors, etc. A 32-bit (or higher) processor is preferred to compute the algorithm.
The Mbed HDK supports onboard components and off board components, allows flexible rapid prototyping. A wireless communication link between the device and a smart phone, computer, or other readout device is supported. It supports Wi-Fi, Bluetooth, and 3G communications, which are commonly available on both computers and smart phones.
The first assessment of the MCU was to check its potential to execute the required calculations in the time available between incoming samples. The most straightforward and quickest approach to test this was to measure how much time does it takes to perform a specific task. The main core of the computation is performing repetitive convolution on the input signal. The signal is filtered by multiple filters and specific features are weighted and combined to generate a SV value. Therefore, the first assessment was to measure the MCU time span required to compute the twelve different filters.
Acceleration data was collected from the MCU and the necessary filtering computed. 3.5 seconds of data were collected, and the MCU computation time span for the twelve filters obtained. The computation time for those filters took 285 milliseconds, which indicated that approximately 8% of the MCU is utilized. In this case the MCU will be in sleep mode ninety present 90% of the time when performing live computations, and will be able to compute all the required calculations on time. Alternately, the MCU program can be ten times more complicated and demanding before the MCU will have difficulty executing it in the time allotted.
The solution was tested using floating point calculations and with a filter set solution. Therefore, another GA run limiting the chromosome maximum SSS to 16 genes was conducted, resulting in 14 filters; where the longest filter consisted of 350 coefficients. Five seconds of data were collected and the convolution computed on the fly for each sampled data point. Each time the microcontroller sampled a new value (100 Hz sampling rate), it computed all 14 filters. As result, the convolution algorithm is computed on each new data point using previously collected data for calculations. The MCU was configured to zero pad to allow calculation on the initial data points. The MCU output calculations were compared to a Matlab convolution function, to test the accuracy of the computations. Since the convolution is performed every 10 milliseconds on each new data point, the computations are finished at 500 data points (i.e. 5 seconds) and the convolution was not continued on the padded section. The MCU was able to compute all calculations in 6.6 milliseconds, on average, which still permits performance of all necessary computations in the allotted time.
The maximum amount of time it takes MCU to complete the whole computation was evaluated. The wavelet computations are performed in real time on each new second data point. Therefore, at every sixteenth data point, all four decomposition levels are performed. The data is sampled at 100 Hz for five seconds, providing 31 output values. The average computation time was 3.3 milliseconds, providing a window of 6.7 milliseconds for further computations. The first decomposition is performed on six data points and the second decomposition is performed on six output values from the first decomposition. 60% of the MCU RAM and 20% of its flash, were consumed, which does not leave much room to perform any additional computations. Two algorithms (one for heart contraction timing and one for stroke volume) need to be performed on the MCU. Therefore, the second algorithm was separately programmed and tested for MCU performance on this additional algorithm before making any hardware decisions. Note that further optimization may reduce memory footprint, and the two algorithms may run sequentially, and therefore use the same physical memory space at different times.
After the filter computations were verified, the acceleration information values were verified following decimation. If the Low Pass (LP) filter is convolved with the High Pass (HP), a new filter is created and if the input signal is convolved with the new filter and decimated by four, in theory, this should result in the same value as if two wavelet decompositions were performed [174]. The first decomposition is on the input signal providing the approximations of the Low Pass (LP) filter and the second decomposition provides the details of the High Pass (HP) filter. This theory was tested but failed to provide satisfactory results, since the final values from the two approaches did not fully match.
Our second approach to compute an efficient algorithm was to perform the decomposition path for each filter. In this way there are fewer computations since a convolution is performed on every second data point from the input level since each result is decimated by two. For example, the LP filter is computed on every second data point which is equivalent to applying it on the input signal and then decimating the output by two. The same is done on the HP filter where convolution is performed on every second data point of the LP filter output. The wavelet computations are performed in real time on each new second data point.
Therefore, at every sixteenth data point, all four decomposition levels are performed. Samples are taken at 100 Hz for five seconds, providing 31 output values. The average computation time was 3.3 milliseconds, providing a window of 6.7 milliseconds for further computations. The data is zero padded at the initialization stage, and later convolution performed where the number of data points is equal to the filter length. The first decomposition is performed on six data points and the second decomposition is performed on six output values from the first decomposition.
The heat rate detection algorithm is computed by the Continuous Wavelet Transform (CWT). The computations are done using cyclic convolution with the “true” Mother wavelets coefficients. This algorithm has two stages. The first stage is adaptive threshold windowing. This program requires 60% of the flash memory and about 80% of the RAM and does not leave much room to add more computations. To test the full program for detecting heart rate, the calculation was performed for ten seconds and then recorded the input and output signal for four seconds, where the total computation time was fourteen seconds. The average calculation time for the four seconds data was about 4.5 milliseconds, so the program can be computed in the time available between each new data point when data collection is at 100 Hertz.
The SV algorithm was programmed first and modified it to provide the same readings as the Matlab algorithm. The final results showed that the algorithm require 3.5 milliseconds to compute and occupies 60% of the RAM and 20% of the flash memory. The HR algorithm occupies 80% of the RAM and 60% of the flash memory. Both algorithms, therefore, cannot run at the same time on the board since both together exceed the amount of memory available on the Freedom-KL25Z evaluation board. The sub program which merges the output of the SV and HR algorithms requires little computation, but will still increase the amount of memory required by 10% of the available memory. Therefore, a new evaluation board with greater RAM and flash memory would be required to implement the full algorithm.
The drive behind minimizing computation time and filter sets was to create a small device which was portable (i.e. battery operated) and so would have limited computation capabilities. Different hardware types were investigated, including ASIC, FPGA, DSP and MCU, the MCU approach was determined the best fit for this application based on power consumption, acceptable computation power, speed to market, development ease, and feature flexibility. Moreover, an ARM based MCU with high performance and low power consumption, and which offered upwards compatibility, was preferred. An open source development platform was employed since it was tested by name users and supported multiple components allowing for rapid prototyping.
The SV and HR algorithms were tested separately, and both were shown to compute in less than half of the available time on the target MCU. Therefore, both algorithms together could be computed in less than ten milliseconds, allowing a 100 Hz sampling rate. However, the code to implement both algorithms could not fit together on the target MCU, and an alternative target MCU is needed which includes more RAM to hold the entire program.
It has been shown that the FRDM-KL25Z evaluation board is sufficiently fast to make the necessary computations in less than 10 ms, however, it would not be possible to compute both HR and SV algorithms simultaneously due to insufficient memory resources on this board. FRDM-KL46Z is an upper level board in the same family as the FRDM-KL25Z, with a built in 16 bit ADC and draws just 6 mA at full working state. The NXP LPC1768 has a 12 bit, 1 Megasample per second ADC, and draws 42 mA, but it runs at 96 MHz which would allow it to compute the required calculations faster, and then go into sleep mode to save power. A 16+ bit ADC is preferred, but techniques, such as subranging, dithering, and the like, can be used to increase the effective number of bits, especially when the required data acquisition rate is well below the sampling rate.
During initial analysis, it was assumed that observation of respiration rate since would not be possible, due to high-pass analog filtering to decrease “noise” in the frequency range of respiration (i.e. below a few Hz). However, the first integration for finding chest velocity shows that the low respiration frequency was observed to contribute significantly to the signal in the SV analysis. Therefore, the system can readily determine and output respiratory motion parameters, including respiratory rate. Because CO is influenced by breathing, incorporating breathing rate into the CO calculation may significantly improve the accuracy of the CO estimates.
The invention may be used as a method, system or apparatus, as programming codes for performing the stated functions and their equivalents on programmable machines, and the like. The aspects of the invention are intended to be separable, and may be implemented in combination, subcombination, and with various permutations of embodiments. Therefore, the various disclosure herein, including that which is represented by acknowledged prior art, may be combined, subcombined and permuted in accordance with the teachings hereof, without departing from the spirit and scope of the invention.
All references cited herein are expressly incorporated herein by reference in their entirety.
The present application is a Non-provisional of, and claims priority under 35 U.S.C. § 119(e) from, U.S. Provisional Patent Application No. 62/175,686, filed Jun. 15, 2015, the entirety of which is expressly incorporated herein by reference.
| Number | Name | Date | Kind |
|---|---|---|---|
| 4036215 | Doll | Jul 1977 | A |
| 4195643 | Pratt, Jr. | Apr 1980 | A |
| 4519395 | Hrushesky | May 1985 | A |
| 4657025 | Orlando | Apr 1987 | A |
| 4679569 | Lee | Jul 1987 | A |
| 4681098 | Lee | Jul 1987 | A |
| 4738264 | Orlando | Apr 1988 | A |
| 4802486 | Goodman et al. | Feb 1989 | A |
| 4817610 | Lee | Apr 1989 | A |
| 4836215 | Lee | Jun 1989 | A |
| 4838275 | Lee | Jun 1989 | A |
| 4848350 | Lee | Jul 1989 | A |
| 4884578 | Morgenstern | Dec 1989 | A |
| 4889123 | Lee | Dec 1989 | A |
| 4889130 | Lee | Dec 1989 | A |
| 4893633 | Lee | Jan 1990 | A |
| 4895155 | Lee | Jan 1990 | A |
| 4911167 | Corenman et al. | Mar 1990 | A |
| 4926866 | Lee | May 1990 | A |
| 4928692 | Goodman et al. | May 1990 | A |
| 4934372 | Corenman et al. | Jun 1990 | A |
| 5197490 | Steiner et al. | Mar 1993 | A |
| 5270529 | Shudo | Dec 1993 | A |
| 5333649 | Shimokawa et al. | Aug 1994 | A |
| 5334444 | Bhoori et al. | Aug 1994 | A |
| 5445162 | Ives | Aug 1995 | A |
| RE35122 | Corenman et al. | Dec 1995 | E |
| 5507785 | Deno | Apr 1996 | A |
| 5602964 | Barrett | Feb 1997 | A |
| 5649065 | Lo et al. | Jul 1997 | A |
| 5718720 | Prutchi et al. | Feb 1998 | A |
| 5745382 | Vilim et al. | Apr 1998 | A |
| 5782884 | Stotts et al. | Jul 1998 | A |
| 5815198 | Vachtsevanos et al. | Sep 1998 | A |
| 5815608 | Lange et al. | Sep 1998 | A |
| 5845266 | Lupien et al. | Dec 1998 | A |
| 5847952 | Samad | Dec 1998 | A |
| 5940825 | Castelli et al. | Aug 1999 | A |
| 5963929 | Lo | Oct 1999 | A |
| 5964720 | Pelz | Oct 1999 | A |
| 5978788 | Castelli et al. | Nov 1999 | A |
| 5995868 | Dorfmeister et al. | Nov 1999 | A |
| 6012046 | Lupien et al. | Jan 2000 | A |
| 6022322 | Prutchi | Feb 2000 | A |
| 6050950 | Mohler | Apr 2000 | A |
| 6053872 | Mohler | Apr 2000 | A |
| 6091791 | Matsumoto et al. | Jul 2000 | A |
| 6091841 | Rogers et al. | Jul 2000 | A |
| 6098051 | Lupien et al. | Aug 2000 | A |
| 6115488 | Rogers et al. | Sep 2000 | A |
| 6121969 | Jain et al. | Sep 2000 | A |
| 6128346 | Suarez et al. | Oct 2000 | A |
| 6137898 | Broussard et al. | Oct 2000 | A |
| 6152879 | Mohler | Nov 2000 | A |
| 6154705 | McCormack et al. | Nov 2000 | A |
| 6155976 | Sackner et al. | Dec 2000 | A |
| 6167146 | Rogers et al. | Dec 2000 | A |
| 6167155 | Kostrzewski et al. | Dec 2000 | A |
| 6171263 | Sullivan | Jan 2001 | B1 |
| 6179783 | Mohler | Jan 2001 | B1 |
| 6205236 | Rogers et al. | Mar 2001 | B1 |
| 6272479 | Farry et al. | Aug 2001 | B1 |
| 6334219 | Hill et al. | Dec 2001 | B1 |
| 6361501 | Amano | Mar 2002 | B1 |
| 6363350 | Lafe | Mar 2002 | B1 |
| 6370424 | Prutchi | Apr 2002 | B1 |
| 6370481 | Gamble | Apr 2002 | B1 |
| 6377306 | Johnson et al. | Apr 2002 | B1 |
| 6389157 | Rogers et al. | May 2002 | B2 |
| 6397136 | Breed et al. | May 2002 | B1 |
| 6400996 | Hoffberg et al. | Jun 2002 | B1 |
| 6434583 | Dapper et al. | Aug 2002 | B1 |
| 6445988 | Breed et al. | Sep 2002 | B1 |
| 6452870 | Breed et al. | Sep 2002 | B1 |
| 6459973 | Breed et al. | Oct 2002 | B1 |
| 6477406 | Turcott | Nov 2002 | B1 |
| 6478744 | Mohler | Nov 2002 | B2 |
| 6510406 | Marchisio | Jan 2003 | B1 |
| 6529809 | Breed et al. | Mar 2003 | B1 |
| 6535644 | Kurapati | Mar 2003 | B1 |
| 6535754 | Fishbein et al. | Mar 2003 | B2 |
| 6547743 | Brydon | Apr 2003 | B2 |
| 6549804 | Osorio et al. | Apr 2003 | B1 |
| 6551252 | Sackner et al. | Apr 2003 | B2 |
| 6556699 | Rogers et al. | Apr 2003 | B2 |
| 6560542 | Mandell et al. | May 2003 | B1 |
| 6601051 | Lo et al. | Jul 2003 | B1 |
| 6640145 | Hoffberg et al. | Oct 2003 | B2 |
| 6647289 | Prutchi | Nov 2003 | B2 |
| 6650766 | Rogers et al. | Nov 2003 | B1 |
| 6650779 | Vachtesvanos et al. | Nov 2003 | B2 |
| 6658287 | Litt et al. | Dec 2003 | B1 |
| 6675164 | Kamath et al. | Jan 2004 | B2 |
| 6678548 | Echauz et al. | Jan 2004 | B1 |
| 6697661 | Raghavan et al. | Feb 2004 | B2 |
| 6708163 | Kargupta et al. | Mar 2004 | B1 |
| 6757415 | Rogers et al. | Jun 2004 | B1 |
| 6757602 | Breed et al. | Jun 2004 | B2 |
| 6763128 | Rogers et al. | Jul 2004 | B1 |
| 6763322 | Potyrailo et al. | Jul 2004 | B2 |
| 6789054 | Makhlouf | Sep 2004 | B1 |
| 6801645 | Collins et al. | Oct 2004 | B1 |
| 6826300 | Liu et al. | Nov 2004 | B2 |
| 6826428 | Chen et al. | Nov 2004 | B1 |
| 6840907 | Brydon | Jan 2005 | B1 |
| 6850252 | Hoffberg | Feb 2005 | B1 |
| 6862710 | Marchisio | Mar 2005 | B1 |
| 6865492 | Mandell et al. | Mar 2005 | B2 |
| 6879729 | Kamath et al. | Apr 2005 | B2 |
| 6882992 | Werbos | Apr 2005 | B1 |
| 6885975 | Srinivasan et al. | Apr 2005 | B2 |
| 6886008 | Blayvas et al. | Apr 2005 | B2 |
| 6970587 | Rogers | Nov 2005 | B1 |
| 6988093 | Pic et al. | Jan 2006 | B2 |
| 6993378 | Wiederhold et al. | Jan 2006 | B2 |
| 6994675 | Sharrock | Feb 2006 | B2 |
| 6996549 | Zhang et al. | Feb 2006 | B2 |
| 7003403 | Dougherty et al. | Feb 2006 | B1 |
| 7006881 | Hoffberg et al. | Feb 2006 | B1 |
| 7007035 | Kamath et al. | Feb 2006 | B2 |
| 7016885 | Mikhael et al. | Mar 2006 | B1 |
| 7019641 | Lakshmanan et al. | Mar 2006 | B1 |
| 7028015 | Poluzzi et al. | Apr 2006 | B2 |
| 7039462 | Pastore et al. | May 2006 | B2 |
| 7039654 | Eder | May 2006 | B1 |
| 7043293 | Baura | May 2006 | B1 |
| 7051017 | Marchisio | May 2006 | B2 |
| 7077810 | Lange et al. | Jul 2006 | B2 |
| 7079896 | Park et al. | Jul 2006 | B1 |
| 7082572 | Pea et al. | Jul 2006 | B2 |
| 7085401 | Averbuch et al. | Aug 2006 | B2 |
| 7139609 | Min et al. | Nov 2006 | B1 |
| 7147246 | Breed et al. | Dec 2006 | B2 |
| 7149320 | Haykin et al. | Dec 2006 | B2 |
| 7162076 | Liu | Jan 2007 | B2 |
| 7164117 | Breed et al. | Jan 2007 | B2 |
| 7177686 | Turcott | Feb 2007 | B1 |
| 7180943 | Arlid et al. | Feb 2007 | B1 |
| 7190149 | Huff et al. | Mar 2007 | B2 |
| 7194143 | Sowa | Mar 2007 | B2 |
| 7206636 | Turcott | Apr 2007 | B1 |
| 7209787 | DiLorenzo | Apr 2007 | B2 |
| 7228238 | Mandell et al. | Jun 2007 | B2 |
| 7231254 | DiLorenzo | Jun 2007 | B2 |
| 7233882 | Srinivasan et al. | Jun 2007 | B2 |
| 7242984 | DiLorenzo | Jul 2007 | B2 |
| 7243945 | Breed et al. | Jul 2007 | B2 |
| 7248923 | Maile et al. | Jul 2007 | B2 |
| 7269537 | Mattern | Sep 2007 | B1 |
| 7270733 | Wikiel et al. | Sep 2007 | B2 |
| 7277758 | DiLorenzo | Oct 2007 | B2 |
| 7286871 | Cohen | Oct 2007 | B2 |
| 7286964 | Kim | Oct 2007 | B2 |
| 7293002 | Starzyk | Nov 2007 | B2 |
| 7295608 | Reynolds et al. | Nov 2007 | B2 |
| 7308126 | Rogers et al. | Dec 2007 | B2 |
| 7310522 | Geile | Dec 2007 | B2 |
| 7314451 | Halperin et al. | Jan 2008 | B2 |
| 7324036 | Petre et al. | Jan 2008 | B2 |
| 7324851 | DiLorenzo | Jan 2008 | B1 |
| 7333851 | Echauz et al. | Feb 2008 | B2 |
| 7338436 | Snell et al. | Mar 2008 | B1 |
| 7366719 | Shaw | Apr 2008 | B2 |
| 7376553 | Quinn | May 2008 | B2 |
| 7379568 | Movellan et al. | May 2008 | B2 |
| 7383237 | Zhang et al. | Jun 2008 | B2 |
| 7385300 | Huff et al. | Jun 2008 | B2 |
| 7392143 | Jayabalan et al. | Jun 2008 | B2 |
| 7395250 | Aggarwal et al. | Jul 2008 | B1 |
| 7396331 | Mack et al. | Jul 2008 | B2 |
| 7401057 | Eder | Jul 2008 | B2 |
| 7401807 | Breed et al. | Jul 2008 | B2 |
| 7403820 | DiLorenzo | Jul 2008 | B2 |
| 7407029 | Breed et al. | Aug 2008 | B2 |
| 7408486 | Kilbank | Aug 2008 | B2 |
| 7409303 | Yeo et al. | Aug 2008 | B2 |
| 7415126 | Breed et al. | Aug 2008 | B2 |
| 7417536 | Lakshmanan et al. | Aug 2008 | B2 |
| 7426499 | Eder | Sep 2008 | B2 |
| 7428323 | Hillman | Sep 2008 | B2 |
| 7430483 | Wiig | Sep 2008 | B2 |
| 7451005 | Hoffberg et al. | Nov 2008 | B2 |
| 7454244 | Kassab et al. | Nov 2008 | B2 |
| 7460903 | Pineda et al. | Dec 2008 | B2 |
| 7460909 | Koh et al. | Dec 2008 | B1 |
| 7477758 | Piirainen et al. | Jan 2009 | B2 |
| 7483868 | Meng et al. | Jan 2009 | B2 |
| 7494459 | Anstadt et al. | Feb 2009 | B2 |
| 7502677 | Weichenberger et al. | Mar 2009 | B2 |
| 7511833 | Breed | Mar 2009 | B2 |
| 7519476 | Tnacheri et al. | Apr 2009 | B1 |
| 7526461 | Srinivasa et al. | Apr 2009 | B2 |
| 7533006 | Huddleston et al. | May 2009 | B2 |
| 7535822 | Geile et al. | May 2009 | B2 |
| 7536064 | Venkatesan et al. | May 2009 | B2 |
| 7539532 | Tran | May 2009 | B2 |
| 7539533 | Tran | May 2009 | B2 |
| 7546161 | Bjorling et al. | Jun 2009 | B1 |
| 7558622 | Tran | Jul 2009 | B2 |
| 7574255 | Min | Aug 2009 | B1 |
| 7575171 | Lev | Aug 2009 | B2 |
| 7577478 | Kroll et al. | Aug 2009 | B1 |
| 7581434 | Discenzo et al. | Sep 2009 | B1 |
| 7584075 | Kim | Sep 2009 | B2 |
| 7587069 | Movellan et al. | Sep 2009 | B2 |
| 7590510 | Kim | Sep 2009 | B2 |
| 7590589 | Hoffberg | Sep 2009 | B2 |
| 7596242 | Breed et al. | Sep 2009 | B2 |
| 7596470 | Kim | Sep 2009 | B2 |
| 7599759 | Zugibe et al. | Oct 2009 | B2 |
| 7604956 | Drukier | Oct 2009 | B2 |
| 7606790 | Levy | Oct 2009 | B2 |
| 7620527 | Gielis | Nov 2009 | B1 |
| 7624076 | Movellan et al. | Nov 2009 | B2 |
| 7630757 | Dorfmeister et al. | Dec 2009 | B2 |
| 7636600 | Koh | Dec 2009 | B1 |
| 7636700 | Owechko et al. | Dec 2009 | B2 |
| 7649160 | Colomb et al. | Jan 2010 | B2 |
| 7650189 | Park et al. | Jan 2010 | B1 |
| 7655895 | Breed | Feb 2010 | B2 |
| 7657299 | Huizenga et al. | Feb 2010 | B2 |
| 7660437 | Breed | Feb 2010 | B2 |
| 7660632 | Kirby et al. | Feb 2010 | B2 |
| 7662785 | DeNardo et al. | Feb 2010 | B2 |
| 7663502 | Breed | Feb 2010 | B2 |
| 7668697 | Volkov et al. | Feb 2010 | B2 |
| 7670295 | Sackner et al. | Mar 2010 | B2 |
| 7672219 | Geile | Mar 2010 | B2 |
| 7672728 | Libbus | Mar 2010 | B2 |
| 7675843 | Geile | Mar 2010 | B2 |
| 7676062 | Breed et al. | Mar 2010 | B2 |
| 7676263 | Harris et al. | Mar 2010 | B2 |
| 7676266 | Kroll | Mar 2010 | B1 |
| 7689283 | Schecter | Mar 2010 | B1 |
| 7693683 | Ihara | Apr 2010 | B2 |
| 7697453 | Geile et al. | Apr 2010 | B2 |
| 7697792 | Keating et al. | Apr 2010 | B2 |
| 7702185 | Keating et al. | Apr 2010 | B2 |
| 7702555 | Breeden et al. | Apr 2010 | B1 |
| 7706349 | Geile et al. | Apr 2010 | B2 |
| 7710828 | Barger et al. | May 2010 | B2 |
| 7712898 | Abramoff et al. | May 2010 | B2 |
| 7715894 | Dunseath et al. | May 2010 | B2 |
| 7716148 | Meng et al. | May 2010 | B2 |
| 7727157 | Sharrock | Jun 2010 | B2 |
| 7730063 | Eder | Jun 2010 | B2 |
| 7733224 | Tran | Jun 2010 | B2 |
| 7742806 | Sternickel et al. | Jun 2010 | B2 |
| 7747325 | Dilorenzo | Jun 2010 | B2 |
| 7747390 | Miyake et al. | Jun 2010 | B2 |
| 7756060 | Geile et al. | Jul 2010 | B2 |
| 7768380 | Breed et al. | Aug 2010 | B2 |
| 7769446 | Moffitt et al. | Aug 2010 | B2 |
| 7769513 | Breed et al. | Aug 2010 | B2 |
| 7773537 | Geile et al. | Aug 2010 | B2 |
| 7774055 | Min | Aug 2010 | B1 |
| 7777743 | Pao et al. | Aug 2010 | B2 |
| 7785257 | Mack et al. | Aug 2010 | B2 |
| 7787946 | Stahmann et al. | Aug 2010 | B2 |
| 7788212 | Beckmann et al. | Aug 2010 | B2 |
| RE41771 | Geile et al. | Sep 2010 | E |
| 7797050 | Libbus et al. | Sep 2010 | B2 |
| 7805386 | Greer | Sep 2010 | B2 |
| 7813805 | Farazi | Oct 2010 | B1 |
| 7813822 | Hoffberg | Oct 2010 | B1 |
| 7818053 | Kassab | Oct 2010 | B2 |
| 7819003 | Breed et al. | Oct 2010 | B2 |
| 7823058 | Pea et al. | Oct 2010 | B2 |
| 7831358 | Breed et al. | Nov 2010 | B2 |
| 7846104 | MacQuarrie et al. | Dec 2010 | B2 |
| 7848816 | Wenzel et al. | Dec 2010 | B1 |
| 7850616 | Gill et al. | Dec 2010 | B1 |
| 7855977 | Morrison et al. | Dec 2010 | B2 |
| 7856268 | Kroll et al. | Dec 2010 | B2 |
| 7869869 | Farazi | Jan 2011 | B1 |
| 7869876 | Prakash et al. | Jan 2011 | B2 |
| 7872985 | Geile et al. | Jan 2011 | B2 |
| 7877146 | Rezai et al. | Jan 2011 | B2 |
| 7881180 | Geile et al. | Feb 2011 | B2 |
| 7881181 | Dapper et al. | Feb 2011 | B2 |
| 7887089 | Breed et al. | Feb 2011 | B2 |
| RE42236 | Geile et al. | Mar 2011 | E |
| 7903617 | Liu et al. | Mar 2011 | B2 |
| 7904187 | Hoffberg et al. | Mar 2011 | B2 |
| 7908004 | Gill et al. | Mar 2011 | B1 |
| 7912138 | Geile et al. | Mar 2011 | B2 |
| 7912734 | Kil | Mar 2011 | B2 |
| 7936662 | Geile et al. | May 2011 | B2 |
| 7949398 | Wenzel et al. | May 2011 | B1 |
| 7949399 | Wenzel et al. | May 2011 | B2 |
| 7953479 | Wenzel et al. | May 2011 | B2 |
| 7957265 | Geile et al. | Jun 2011 | B2 |
| 7963925 | Schecter | Jun 2011 | B1 |
| 7966078 | Hoffberg et al. | Jun 2011 | B2 |
| 7974714 | Hoffberg | Jul 2011 | B2 |
| 7981399 | Burns | Jul 2011 | B2 |
| 7983141 | Geile | Jul 2011 | B2 |
| 7983817 | Breed | Jul 2011 | B2 |
| 7987003 | Hoffberg et al. | Jul 2011 | B2 |
| 7995454 | Geile et al. | Aug 2011 | B2 |
| 7996762 | Qi et al. | Aug 2011 | B2 |
| 8000773 | Rousso et al. | Aug 2011 | B2 |
| 8005524 | Brauker et al. | Aug 2011 | B2 |
| 8005543 | Libbus et al. | Aug 2011 | B2 |
| 8005631 | Barger et al. | Aug 2011 | B2 |
| 8010194 | Muller | Aug 2011 | B2 |
| 8015128 | Baughman | Sep 2011 | B2 |
| 8023710 | Summers et al. | Sep 2011 | B2 |
| 8024044 | Kirby et al. | Sep 2011 | B2 |
| 8025624 | Wariar et al. | Sep 2011 | B2 |
| 8027724 | Wei et al. | Sep 2011 | B2 |
| 8031060 | Hoffberg et al. | Oct 2011 | B2 |
| 8032477 | Hoffberg et al. | Oct 2011 | B1 |
| 8036265 | Reynolds et al. | Oct 2011 | B1 |
| 8036442 | Bernard et al. | Oct 2011 | B2 |
| 8036736 | Snyder et al. | Oct 2011 | B2 |
| 8041124 | Shi et al. | Oct 2011 | B2 |
| 8041651 | Greer | Oct 2011 | B2 |
| 8046107 | Zugibe et al. | Oct 2011 | B2 |
| 8055667 | Levy | Nov 2011 | B2 |
| 8068894 | Huizenga et al. | Nov 2011 | B2 |
| 8073541 | Alt et al. | Dec 2011 | B2 |
| 8077958 | Qian et al. | Dec 2011 | B2 |
| 8078274 | Kassab | Dec 2011 | B2 |
| 8078552 | Kaushal et al. | Dec 2011 | B2 |
| 8082032 | Kassab et al. | Dec 2011 | B2 |
| 8082353 | Huber et al. | Dec 2011 | B2 |
| 8086017 | He et al. | Dec 2011 | B2 |
| 8086294 | Echauz et al. | Dec 2011 | B2 |
| 8089853 | Dapper et al. | Jan 2012 | B2 |
| 8094551 | Huber et al. | Jan 2012 | B2 |
| 8098938 | Buddemeier et al. | Jan 2012 | B1 |
| 8099161 | Kassab | Jan 2012 | B2 |
| 8103333 | Tran | Jan 2012 | B2 |
| 8107726 | Xu et al. | Jan 2012 | B2 |
| 8108036 | Tran | Jan 2012 | B2 |
| 8114143 | Kassab et al. | Feb 2012 | B2 |
| 8115640 | Walls | Feb 2012 | B2 |
| 8121046 | Morrison et al. | Feb 2012 | B2 |
| 8121673 | Tran | Feb 2012 | B2 |
| 8121823 | Krebs et al. | Feb 2012 | B2 |
| 8126664 | Fournier et al. | Feb 2012 | B2 |
| 8145304 | Moffitt et al. | Mar 2012 | B2 |
| 8149649 | Brinn et al. | Apr 2012 | B1 |
| 8160701 | Zhao et al. | Apr 2012 | B2 |
| 8164345 | Bushnell et al. | Apr 2012 | B2 |
| 8165661 | Koski et al. | Apr 2012 | B2 |
| 8165916 | Hoffberg et al. | Apr 2012 | B2 |
| 8170333 | Balster et al. | May 2012 | B2 |
| 8170334 | Balster et al. | May 2012 | B2 |
| 8170335 | Balster et al. | May 2012 | B2 |
| 8174956 | Dapper et al. | May 2012 | B2 |
| 8179847 | Huber et al. | May 2012 | B2 |
| 8183062 | Funk et al. | May 2012 | B2 |
| 8185194 | Kassab | May 2012 | B2 |
| 8190194 | Brisebois | May 2012 | B2 |
| 8190543 | Kaushal et al. | May 2012 | B2 |
| 8194938 | Wechsler et al. | Jun 2012 | B2 |
| 8194986 | Conwell | Jun 2012 | B2 |
| 8199632 | Geile et al. | Jun 2012 | B2 |
| 8200506 | Kil | Jun 2012 | B2 |
| 8204697 | Garvey et al. | Jun 2012 | B2 |
| 8208697 | Schaffer et al. | Jun 2012 | B2 |
| 8209010 | Ryu et al. | Jun 2012 | B2 |
| 8209745 | Huber et al. | Jun 2012 | B2 |
| 8213398 | Geile et al. | Jul 2012 | B2 |
| 8213399 | Geile et al. | Jul 2012 | B2 |
| 8214033 | Snell et al. | Jul 2012 | B2 |
| 8216139 | Brauker et al. | Jul 2012 | B2 |
| 8223023 | Sachanandani et al. | Jul 2012 | B2 |
| 8226561 | McLaughlin et al. | Jul 2012 | B2 |
| 8233958 | Brauker et al. | Jul 2012 | B2 |
| 8244355 | Bennett et al. | Aug 2012 | B2 |
| 8244475 | Aguilar et al. | Aug 2012 | B2 |
| 8249698 | Mugler et al. | Aug 2012 | B2 |
| 8251906 | Brauker et al. | Aug 2012 | B2 |
| 8251911 | MacQuarrie et al. | Aug 2012 | B2 |
| 8253824 | Benezra et al. | Aug 2012 | B2 |
| 8255042 | MacQuarrie et al. | Aug 2012 | B2 |
| 8257259 | Brauker et al. | Sep 2012 | B2 |
| 8262582 | Kortelainen | Sep 2012 | B2 |
| 8265725 | Brauker et al. | Sep 2012 | B2 |
| 8271412 | Johnson | Sep 2012 | B2 |
| 8275463 | Andersen | Sep 2012 | B1 |
| 8282549 | Brauker et al. | Oct 2012 | B2 |
| 8285373 | Ternes et al. | Oct 2012 | B2 |
| 8287459 | Min et al. | Oct 2012 | B2 |
| 8290561 | Brauker et al. | Oct 2012 | B2 |
| 8295934 | Leyde | Oct 2012 | B2 |
| 8301241 | Ternes et al. | Oct 2012 | B2 |
| 8301406 | Lee et al. | Oct 2012 | B2 |
| 8305436 | Fujisawa et al. | Nov 2012 | B2 |
| 8307273 | Pea et al. | Nov 2012 | B2 |
| 8315150 | Geile et al. | Nov 2012 | B2 |
| 8320217 | Barger et al. | Nov 2012 | B1 |
| 8321003 | Zhang et al. | Nov 2012 | B2 |
| 8323188 | Tran | Dec 2012 | B2 |
| 8323189 | Tran et al. | Dec 2012 | B2 |
| 8326428 | Wenzel et al. | Dec 2012 | B2 |
| 8326429 | Wenzel et al. | Dec 2012 | B2 |
| 8328718 | Tran | Dec 2012 | B2 |
| 8331228 | Huber et al. | Dec 2012 | B2 |
| 8346360 | Libbus et al. | Jan 2013 | B2 |
| 8351321 | Dapper et al. | Jan 2013 | B2 |
| 8355579 | LeMoigne-Stewart et al. | Jan 2013 | B2 |
| 8364136 | Hoffberg et al. | Jan 2013 | B2 |
| 8366707 | Kassab et al. | Feb 2013 | B2 |
| 8369967 | Hoffberg et al. | Feb 2013 | B2 |
| 8374414 | Tang et al. | Feb 2013 | B2 |
| 8374667 | Brauker et al. | Feb 2013 | B2 |
| 8374696 | Sanchez et al. | Feb 2013 | B2 |
| 8376954 | Lange et al. | Feb 2013 | B2 |
| 8376964 | Park et al. | Feb 2013 | B2 |
| 8380308 | Rosenberg et al. | Feb 2013 | B2 |
| 8388604 | Kassab | Mar 2013 | B2 |
| 8391963 | Sternickel et al. | Mar 2013 | B2 |
| 8396582 | Kaushal et al. | Mar 2013 | B2 |
| 8397204 | Sengupta et al. | Mar 2013 | B2 |
| 8401640 | Zhao et al. | Mar 2013 | B2 |
| 8403865 | Halperin et al. | Mar 2013 | B2 |
| 8406115 | Dapper et al. | Mar 2013 | B2 |
| 8406522 | Owechko et al. | Mar 2013 | B1 |
| 8406867 | Kassab | Mar 2013 | B2 |
| 8411910 | Savvides et al. | Apr 2013 | B2 |
| 8416710 | Brisebois et al. | Apr 2013 | B2 |
| 8423125 | Rousso et al. | Apr 2013 | B2 |
| 8425415 | Tran | Apr 2013 | B2 |
| 8427649 | Hays et al. | Apr 2013 | B2 |
| 8428925 | Krebs et al. | Apr 2013 | B2 |
| 8433101 | Xu et al. | Apr 2013 | B2 |
| 8435167 | Oohashi et al. | May 2013 | B2 |
| 8435179 | Goode, Jr. et al. | May 2013 | B2 |
| 8437223 | Barger et al. | May 2013 | B2 |
| 8437844 | Syed Momen et al. | May 2013 | B2 |
| 8437998 | Routh et al. | May 2013 | B2 |
| 8445851 | Rousso et al. | May 2013 | B2 |
| 8449471 | Tran | May 2013 | B2 |
| 8452389 | Min | May 2013 | B2 |
| 8456309 | Sachanandani et al. | Jun 2013 | B2 |
| 8461988 | Tran | Jun 2013 | B2 |
| 8463441 | Zugibe et al. | Jun 2013 | B2 |
| 8467611 | Kumar et al. | Jun 2013 | B2 |
| 8467874 | Chen et al. | Jun 2013 | B2 |
| 8467884 | Chen et al. | Jun 2013 | B2 |
| 8469886 | Brauker et al. | Jun 2013 | B2 |
| 8473055 | Ryu et al. | Jun 2013 | B2 |
| 8473068 | Farazi | Jun 2013 | B2 |
| 8475367 | Yuen et al. | Jul 2013 | B1 |
| 8475368 | Tran et al. | Jul 2013 | B2 |
| 8478394 | Prichep et al. | Jul 2013 | B2 |
| 8482416 | Krans et al. | Jul 2013 | B2 |
| 8486690 | Burns | Jul 2013 | B2 |
| 8488863 | Boucheron | Jul 2013 | B2 |
| 8489247 | Engler | Jul 2013 | B1 |
| 8491492 | Shinar et al. | Jul 2013 | B2 |
| 8494829 | Teixeira | Jul 2013 | B2 |
| RE44460 | Geile | Aug 2013 | E |
| 8500636 | Tran | Aug 2013 | B2 |
| 8503791 | Conwell | Aug 2013 | B2 |
| 8509881 | Thiagarajan et al. | Aug 2013 | B2 |
| 8509890 | Keel et al. | Aug 2013 | B2 |
| 8515126 | Medasani et al. | Aug 2013 | B1 |
| 8515534 | Ternes et al. | Aug 2013 | B2 |
| 8515535 | Hopper et al. | Aug 2013 | B2 |
| 8516266 | Hoffberg et al. | Aug 2013 | B2 |
| 8517953 | Lange et al. | Aug 2013 | B2 |
| 8520979 | Conwell | Aug 2013 | B2 |
| 8522312 | Huber et al. | Aug 2013 | B2 |
| 8525673 | Tran | Sep 2013 | B2 |
| 8525687 | Tran | Sep 2013 | B2 |
| 8527324 | Richter | Sep 2013 | B2 |
| 8531291 | Tran | Sep 2013 | B2 |
| 8532770 | Sweeney et al. | Sep 2013 | B2 |
| 8536133 | DeNardo et al. | Sep 2013 | B2 |
| 8540651 | Pfeffer et al. | Sep 2013 | B2 |
| 8543199 | Snyder et al. | Sep 2013 | B2 |
| 8547824 | Dapper et al. | Oct 2013 | B2 |
| 8559645 | Corona-Strauss et al. | Oct 2013 | B2 |
| 8560134 | Lee | Oct 2013 | B1 |
| 8562526 | Heneghan et al. | Oct 2013 | B2 |
| 8576693 | Dapper et al. | Nov 2013 | B2 |
| 8577451 | Causevic | Nov 2013 | B2 |
| 8577822 | Kusiak et al. | Nov 2013 | B2 |
| 8582916 | Bar-Aviv et al. | Nov 2013 | B2 |
| 8583229 | Rezai et al. | Nov 2013 | B2 |
| 8583230 | Ryu et al. | Nov 2013 | B2 |
| 8583233 | Betzold | Nov 2013 | B2 |
| 8583263 | Hoffberg et al. | Nov 2013 | B2 |
| 8585607 | Klap et al. | Nov 2013 | B2 |
| 8586932 | Rousso et al. | Nov 2013 | B2 |
| 8588906 | Ternes et al. | Nov 2013 | B2 |
| 8588933 | Floyd et al. | Nov 2013 | B2 |
| 8594787 | Crompvoets et al. | Nov 2013 | B2 |
| 8594805 | Hincapie Ordonez et al. | Nov 2013 | B2 |
| 8594811 | Chen et al. | Nov 2013 | B2 |
| 8595164 | Dong et al. | Nov 2013 | B2 |
| 8599266 | Trivedi et al. | Dec 2013 | B2 |
| 8600830 | Hoffberg | Dec 2013 | B2 |
| 8603010 | Lange et al. | Dec 2013 | B2 |
| 8605970 | Bar-Aviv et al. | Dec 2013 | B2 |
| 8606021 | Conwell | Dec 2013 | B2 |
| 8606349 | Rousso et al. | Dec 2013 | B2 |
| 8606418 | Myers et al. | Dec 2013 | B1 |
| 8611692 | Nimnual et al. | Dec 2013 | B2 |
| 8617082 | Zhang et al. | Dec 2013 | B2 |
| 8620426 | Moffitt et al. | Dec 2013 | B2 |
| 8620427 | Libbus et al. | Dec 2013 | B2 |
| 8620660 | Yen et al. | Dec 2013 | B2 |
| 8626274 | Chiu et al. | Jan 2014 | B2 |
| 8626281 | Ternes et al. | Jan 2014 | B2 |
| 8630707 | Zhao et al. | Jan 2014 | B2 |
| 8632469 | Kassab | Jan 2014 | B2 |
| 8634930 | Dalal et al. | Jan 2014 | B2 |
| 8635051 | Wu et al. | Jan 2014 | B1 |
| 8638655 | Geile | Jan 2014 | B2 |
| 8645832 | Pea et al. | Feb 2014 | B2 |
| 8648959 | Capata et al. | Feb 2014 | B2 |
| 8649565 | Kim et al. | Feb 2014 | B1 |
| 8649853 | Sweeney et al. | Feb 2014 | B2 |
| 8652038 | Tran et al. | Feb 2014 | B2 |
| 8657745 | Brauker et al. | Feb 2014 | B2 |
| 8659697 | Capata et al. | Feb 2014 | B2 |
| 8677505 | Redlich et al. | Mar 2014 | B2 |
| 8678943 | Mooney | Mar 2014 | B2 |
| 8679018 | McLaughlin et al. | Mar 2014 | B2 |
| 8679030 | Shinar et al. | Mar 2014 | B2 |
| 8679034 | Halperin et al. | Mar 2014 | B2 |
| 8680991 | Tran | Mar 2014 | B2 |
| 8684900 | Tran | Apr 2014 | B2 |
| 8684922 | Tran | Apr 2014 | B2 |
| 8684925 | Manicka et al. | Apr 2014 | B2 |
| 8696569 | Yuen et al. | Apr 2014 | B2 |
| 8700137 | Albert | Apr 2014 | B2 |
| 8700146 | Ternes et al. | Apr 2014 | B2 |
| 8702616 | Gill et al. | Apr 2014 | B2 |
| 8702629 | Giuffrida et al. | Apr 2014 | B2 |
| 8706204 | Seo et al. | Apr 2014 | B2 |
| 8708903 | Tran | Apr 2014 | B2 |
| 8712507 | Cazares et al. | Apr 2014 | B2 |
| 8713025 | Eder | Apr 2014 | B2 |
| 8714983 | Kil | May 2014 | B2 |
| 8725243 | Dilorenzo et al. | May 2014 | B2 |
| 8725507 | Cortez et al. | May 2014 | B2 |
| 8725667 | Kaushal et al. | May 2014 | B2 |
| 8727978 | Tran et al. | May 2014 | B2 |
| 8731646 | Halperin et al. | May 2014 | B2 |
| 8734360 | Klap et al. | May 2014 | B2 |
| 8738111 | Sweeney et al. | May 2014 | B2 |
| 8738119 | Zhang et al. | May 2014 | B2 |
| 8743776 | Gurajala et al. | Jun 2014 | B2 |
| 8744557 | Sternickel et al. | Jun 2014 | B2 |
| 8744607 | Kaushal et al. | Jun 2014 | B2 |
| 8747312 | Yuen et al. | Jun 2014 | B2 |
| 8747313 | Tran et al. | Jun 2014 | B2 |
| 8747315 | Brauker et al. | Jun 2014 | B2 |
| 8747336 | Tran | Jun 2014 | B2 |
| 8750971 | Tran | Jun 2014 | B2 |
| 8750981 | Bornzin et al. | Jun 2014 | B2 |
| 8750992 | Hopper et al. | Jun 2014 | B2 |
| 8755837 | Rhoads et al. | Jun 2014 | B2 |
| 8755916 | Lou | Jun 2014 | B2 |
| 8755940 | Lou et al. | Jun 2014 | B2 |
| 8761051 | Brisebois et al. | Jun 2014 | B2 |
| 8761893 | Klefenz et al. | Jun 2014 | B2 |
| 8761903 | Chen et al. | Jun 2014 | B2 |
| 8762065 | DiLorenzo | Jun 2014 | B2 |
| 8764651 | Tran | Jul 2014 | B2 |
| 8764653 | Kaminska et al. | Jul 2014 | B2 |
| 8764661 | McLaughlin et al. | Jul 2014 | B2 |
| 8768461 | Stein | Jul 2014 | B2 |
| 8775134 | Gielis | Jul 2014 | B2 |
| 8775143 | Routh et al. | Jul 2014 | B2 |
| 8781597 | DiLorenzo | Jul 2014 | B2 |
| 8786624 | Echauz et al. | Jul 2014 | B2 |
| 8790264 | Sandler et al. | Jul 2014 | B2 |
| 8792974 | Rothman | Jul 2014 | B2 |
| 8792998 | Xi et al. | Jul 2014 | B2 |
| 8797448 | Capata et al. | Aug 2014 | B2 |
| 8797550 | Hays et al. | Aug 2014 | B2 |
| 8798714 | Henning | Aug 2014 | B2 |
| 8798726 | Kang et al. | Aug 2014 | B2 |
| 8798731 | Keel et al. | Aug 2014 | B2 |
| 8801610 | Brauker et al. | Aug 2014 | B2 |
| 8810796 | Hays et al. | Aug 2014 | B2 |
| 8811977 | Austin et al. | Aug 2014 | B2 |
| 8818404 | Brisebois | Aug 2014 | B2 |
| 8818524 | Hincapie Ordonez et al. | Aug 2014 | B2 |
| 8818778 | Salazar-Tio et al. | Aug 2014 | B2 |
| 8821403 | Sharrock | Sep 2014 | B2 |
| 8821418 | Meger et al. | Sep 2014 | B2 |
| 8826199 | Sedaghat et al. | Sep 2014 | B2 |
| 8827918 | Kim et al. | Sep 2014 | B2 |
| 8831705 | Dobak | Sep 2014 | B2 |
| 8834364 | Heneghan et al. | Sep 2014 | B2 |
| 8838510 | Baughman et al. | Sep 2014 | B2 |
| 8840564 | Pinhas et al. | Sep 2014 | B2 |
| 8842136 | Kuo et al. | Sep 2014 | B2 |
| 8849390 | Echauz et al. | Sep 2014 | B2 |
| 8849575 | Gustafsson et al. | Sep 2014 | B2 |
| 8849629 | Brown et al. | Sep 2014 | B2 |
| 8849737 | Engler | Sep 2014 | B1 |
| 8850048 | Huber et al. | Sep 2014 | B2 |
| 8855011 | Ortega et al. | Oct 2014 | B2 |
| 8855372 | Rodriguez et al. | Oct 2014 | B2 |
| 8855775 | Leyde | Oct 2014 | B2 |
| 8855980 | Brown et al. | Oct 2014 | B2 |
| 8856716 | Huang et al. | Oct 2014 | B1 |
| 8858449 | Inan et al. | Oct 2014 | B2 |
| 8860793 | Lo et al. | Oct 2014 | B2 |
| 8861799 | Savvides et al. | Oct 2014 | B2 |
| 8862226 | Ternes et al. | Oct 2014 | B2 |
| 8862231 | Kirchner et al. | Oct 2014 | B2 |
| 8862627 | Ferringer et al. | Oct 2014 | B2 |
| 8866322 | Tchoryk, Jr. et al. | Oct 2014 | B2 |
| 8866936 | Barbu | Oct 2014 | B2 |
| 8868172 | Leyde et al. | Oct 2014 | B2 |
| 8870780 | Inan et al. | Oct 2014 | B2 |
| 8873813 | Tadayon et al. | Oct 2014 | B2 |
| 8873853 | Rodriguez | Oct 2014 | B2 |
| 8874203 | Kassab et al. | Oct 2014 | B2 |
| 8874477 | Hoffberg | Oct 2014 | B2 |
| 8880156 | Crompvoets et al. | Nov 2014 | B2 |
| 8880158 | Spector | Nov 2014 | B2 |
| 8882684 | Halperin et al. | Nov 2014 | B2 |
| 8882765 | Kassab et al. | Nov 2014 | B2 |
| 8886301 | Kassab | Nov 2014 | B2 |
| 8886311 | Anderson et al. | Nov 2014 | B2 |
| 8888710 | Wariar et al. | Nov 2014 | B2 |
| 8892413 | Routh et al. | Nov 2014 | B2 |
| 8897586 | Dorairaj | Nov 2014 | B2 |
| 8897869 | Kassab | Nov 2014 | B2 |
| 8903997 | Hobbs | Dec 2014 | B2 |
| 8905928 | Hayes et al. | Dec 2014 | B2 |
| 8909329 | Prakash et al. | Dec 2014 | B2 |
| 8913839 | Ricanek, Jr. et al. | Dec 2014 | B2 |
| 8918169 | Kassab et al. | Dec 2014 | B2 |
| 8918172 | Moffitt et al. | Dec 2014 | B2 |
| 8922856 | Yanai et al. | Dec 2014 | B2 |
| 8923958 | Gupta et al. | Dec 2014 | B2 |
| 8923965 | Min et al. | Dec 2014 | B2 |
| 8923981 | Grill, Jr. et al. | Dec 2014 | B2 |
| 8934970 | Ternes et al. | Jan 2015 | B2 |
| 8938113 | Kovalan et al. | Jan 2015 | B2 |
| 8942180 | Gurajala et al. | Jan 2015 | B2 |
| 8942779 | Halperin et al. | Jan 2015 | B2 |
| 8942799 | Ternes et al. | Jan 2015 | B2 |
| 8945875 | Roessler et al. | Feb 2015 | B2 |
| 8948442 | Breed et al. | Feb 2015 | B2 |
| 8954146 | Hopper et al. | Feb 2015 | B2 |
| 8961185 | Bleich et al. | Feb 2015 | B2 |
| 8965044 | Owechko et al. | Feb 2015 | B1 |
| 8968195 | Tran | Mar 2015 | B2 |
| 8972861 | Pea et al. | Mar 2015 | B2 |
| 8976856 | Haskell | Mar 2015 | B2 |
| 8979730 | Naujokat et al. | Mar 2015 | B2 |
| 8979731 | Aarts et al. | Mar 2015 | B2 |
| 8983854 | Park et al. | Mar 2015 | B2 |
| 8990688 | Lee et al. | Mar 2015 | B2 |
| 8990740 | Zhang et al. | Mar 2015 | B2 |
| 8992434 | Halperin et al. | Mar 2015 | B2 |
| 8995074 | Low | Mar 2015 | B1 |
| 8996107 | Libbus et al. | Mar 2015 | B2 |
| 8996110 | Sison et al. | Mar 2015 | B2 |
| 8996442 | Gould et al. | Mar 2015 | B2 |
| 8998820 | Jarverud et al. | Apr 2015 | B2 |
| 8998830 | Halperin et al. | Apr 2015 | B2 |
| 9002446 | Wenzel et al. | Apr 2015 | B2 |
| 9002453 | Keel et al. | Apr 2015 | B2 |
| 9002483 | Engler | Apr 2015 | B1 |
| 9002682 | Kasabov | Apr 2015 | B2 |
| 9008367 | Tolkowsky et al. | Apr 2015 | B2 |
| 9008754 | Steinberg et al. | Apr 2015 | B2 |
| 9008762 | Brockway et al. | Apr 2015 | B2 |
| 9009670 | Precious et al. | Apr 2015 | B2 |
| 9011338 | Park et al. | Apr 2015 | B2 |
| 9011346 | Wiard et al. | Apr 2015 | B2 |
| 9014453 | Steinberg et al. | Apr 2015 | B2 |
| 9014789 | Mercader et al. | Apr 2015 | B2 |
| 9014790 | Richards et al. | Apr 2015 | B2 |
| 9014809 | Wenzel et al. | Apr 2015 | B2 |
| 9015093 | Commons | Apr 2015 | B1 |
| 9015145 | Lim et al. | Apr 2015 | B2 |
| 9017691 | Barouch et al. | Apr 2015 | B2 |
| 9019819 | Huber et al. | Apr 2015 | B2 |
| 9020585 | John et al. | Apr 2015 | B2 |
| RE45512 | Tearney et al. | May 2015 | E |
| 9022564 | Watanabe et al. | May 2015 | B2 |
| 9023984 | Hutton et al. | May 2015 | B2 |
| 9026193 | Pahlevan et al. | May 2015 | B2 |
| 9026199 | Halperin et al. | May 2015 | B2 |
| 9026200 | Nagata et al. | May 2015 | B2 |
| 9026201 | Zhang et al. | May 2015 | B2 |
| 9026202 | Albert | May 2015 | B2 |
| 9026214 | Ternes et al. | May 2015 | B2 |
| 9028405 | Tran | May 2015 | B2 |
| 9028407 | Bennett-Guerrero | May 2015 | B1 |
| 9029413 | Munger et al. | May 2015 | B2 |
| 9031301 | Imamura et al. | May 2015 | B2 |
| 9031642 | Ghosh | May 2015 | B2 |
| 9031649 | Ousdigian | May 2015 | B2 |
| 9033883 | Wang et al. | May 2015 | B2 |
| 9033887 | Ionasec et al. | May 2015 | B2 |
| 9033893 | Spector | May 2015 | B2 |
| 9037223 | Oral et al. | May 2015 | B2 |
| 9037224 | Fu | May 2015 | B1 |
| 9042619 | Zheng et al. | May 2015 | B2 |
| 9042952 | Lynn et al. | May 2015 | B2 |
| 9042958 | Karmarkar et al. | May 2015 | B2 |
| 9044144 | Figgatt et al. | Jun 2015 | B2 |
| 9044149 | Richards et al. | Jun 2015 | B2 |
| 9044171 | Venkatraman et al. | Jun 2015 | B2 |
| 9044558 | Baker, Jr. et al. | Jun 2015 | B2 |
| 9047272 | To | Jun 2015 | B1 |
| 9047353 | Kumar | Jun 2015 | B2 |
| 9049981 | Patangay et al. | Jun 2015 | B2 |
| 9049985 | Feher | Jun 2015 | B2 |
| 9050007 | Brockway et al. | Jun 2015 | B1 |
| 9050014 | Zhang | Jun 2015 | B2 |
| 9051379 | Wolf et al. | Jun 2015 | B2 |
| 9053416 | De Leo et al. | Jun 2015 | B1 |
| 9055871 | Inan et al. | Jun 2015 | B2 |
| 9055884 | Piferi et al. | Jun 2015 | B2 |
| 9056172 | Shelly et al. | Jun 2015 | B2 |
| 9060669 | Mo et al. | Jun 2015 | B1 |
| 9060683 | Tran | Jun 2015 | B2 |
| 9060689 | Tearney et al. | Jun 2015 | B2 |
| 9060695 | Peters | Jun 2015 | B2 |
| 9060722 | Teixeira | Jun 2015 | B2 |
| 9060733 | Bruder et al. | Jun 2015 | B2 |
| 9063139 | Allikmets et al. | Jun 2015 | B2 |
| 9066672 | Kassab et al. | Jun 2015 | B2 |
| 9066679 | Beach et al. | Jun 2015 | B2 |
| 9066680 | Al-Ali et al. | Jun 2015 | B1 |
| 9066686 | Lasser et al. | Jun 2015 | B2 |
| 9066708 | Kassab | Jun 2015 | B2 |
| 9069130 | Yun et al. | Jun 2015 | B2 |
| 9072437 | Paalasmaa | Jul 2015 | B2 |
| 9072438 | Brockway et al. | Jul 2015 | B2 |
| 9075446 | Garudadri et al. | Jul 2015 | B2 |
| 9076202 | Courtney et al. | Jul 2015 | B2 |
| 9079060 | Hong et al. | Jul 2015 | B2 |
| 9081148 | Tearney et al. | Jul 2015 | B2 |
| 9084531 | Chen et al. | Jul 2015 | B2 |
| 9084576 | Guracar | Jul 2015 | B2 |
| 9084611 | Amirana et al. | Jul 2015 | B2 |
| 9086467 | Elgort et al. | Jul 2015 | B2 |
| 9087368 | Tearney et al. | Jul 2015 | B2 |
| 9089269 | Narayan et al. | Jul 2015 | B2 |
| 9092691 | Beaumont et al. | Jul 2015 | B1 |
| 9095266 | Fu | Aug 2015 | B1 |
| 9095313 | Tolkowsky et al. | Aug 2015 | B2 |
| 9095505 | Hertle | Aug 2015 | B2 |
| 9097756 | Piferi | Aug 2015 | B2 |
| 9101286 | Tolkowsky et al. | Aug 2015 | B2 |
| 9101772 | Kleckner et al. | Aug 2015 | B2 |
| 9107571 | Strauss et al. | Aug 2015 | B2 |
| 9107584 | Wohlschlager et al. | Aug 2015 | B2 |
| 9107586 | Tran | Aug 2015 | B2 |
| 9107623 | Brauker et al. | Aug 2015 | B2 |
| 9107624 | Darty | Aug 2015 | B2 |
| 9113794 | Hong et al. | Aug 2015 | B2 |
| 9113795 | Hong et al. | Aug 2015 | B2 |
| 9113826 | Osumi | Aug 2015 | B2 |
| 9113830 | Galen et al. | Aug 2015 | B2 |
| 9114260 | Pless et al. | Aug 2015 | B2 |
| 9116835 | Smyth | Aug 2015 | B1 |
| RE45660 | Weiner et al. | Sep 2015 | E |
| 9119547 | Cazares et al. | Sep 2015 | B2 |
| 9119573 | Lu et al. | Sep 2015 | B2 |
| 9121801 | Clark et al. | Sep 2015 | B2 |
| 9125574 | Zia et al. | Sep 2015 | B2 |
| 9125577 | Sameni et al. | Sep 2015 | B2 |
| 9125578 | Grunwald | Sep 2015 | B2 |
| 9126050 | Simon et al. | Sep 2015 | B2 |
| 9131852 | Li et al. | Sep 2015 | B2 |
| 9131864 | Korenberg | Sep 2015 | B2 |
| 9131902 | Halperin et al. | Sep 2015 | B2 |
| 9136980 | Baheti et al. | Sep 2015 | B2 |
| 9138150 | Boschetti Sacco et al. | Sep 2015 | B2 |
| 9144394 | Cohen et al. | Sep 2015 | B2 |
| 9147268 | Chen et al. | Sep 2015 | B2 |
| 9148839 | Hejazi et al. | Sep 2015 | B2 |
| RE45725 | Mugler, III et al. | Oct 2015 | E |
| 9149219 | Goode, Jr. et al. | Oct 2015 | B2 |
| 9149231 | Fujita et al. | Oct 2015 | B2 |
| 9149244 | Anderson | Oct 2015 | B2 |
| 9149645 | Sanghera et al. | Oct 2015 | B2 |
| 9155482 | Anderson et al. | Oct 2015 | B2 |
| 9155484 | Baker et al. | Oct 2015 | B2 |
| 9155893 | Zhou et al. | Oct 2015 | B2 |
| 9161705 | Tamil et al. | Oct 2015 | B2 |
| 9162074 | Allavatam et al. | Oct 2015 | B2 |
| 9163216 | Hickman et al. | Oct 2015 | B1 |
| 9164481 | Tsang | Oct 2015 | B2 |
| 9167971 | Ahmad et al. | Oct 2015 | B2 |
| 9168290 | Heifetz | Oct 2015 | B2 |
| 9168380 | Greenhut et al. | Oct 2015 | B1 |
| 9168385 | Snyder | Oct 2015 | B2 |
| 9168419 | Hong et al. | Oct 2015 | B2 |
| 9171250 | Wang et al. | Oct 2015 | B2 |
| 9171353 | Vija et al. | Oct 2015 | B2 |
| 9173566 | Feher | Nov 2015 | B2 |
| 9173579 | Berkow | Nov 2015 | B2 |
| 9173909 | Lee | Nov 2015 | B2 |
| 9174061 | Freeman et al. | Nov 2015 | B2 |
| 9175095 | Deisseroth et al. | Nov 2015 | B2 |
| 9176319 | Bouma et al. | Nov 2015 | B2 |
| 9178330 | Oh et al. | Nov 2015 | B2 |
| 9179890 | Ionasec et al. | Nov 2015 | B2 |
| 9180043 | Merrill et al. | Nov 2015 | B2 |
| 9180300 | Sambelashvili et al. | Nov 2015 | B2 |
| 9183351 | Shusterman | Nov 2015 | B2 |
| 9183626 | Zhao et al. | Nov 2015 | B2 |
| 9186066 | Tearney et al. | Nov 2015 | B2 |
| 9186067 | Tearney et al. | Nov 2015 | B2 |
| 9186068 | Li et al. | Nov 2015 | B2 |
| 9186079 | Mase et al. | Nov 2015 | B2 |
| 9186105 | Leininger et al. | Nov 2015 | B2 |
| 9186106 | Osorio | Nov 2015 | B2 |
| 9186521 | Quan et al. | Nov 2015 | B2 |
| 9189733 | Thompson et al. | Nov 2015 | B2 |
| 9192328 | Brauker et al. | Nov 2015 | B2 |
| 9192336 | Addison et al. | Nov 2015 | B2 |
| 9192446 | Piferi et al. | Nov 2015 | B2 |
| 9193442 | Young et al. | Nov 2015 | B1 |
| 9195949 | Kasabov | Nov 2015 | B2 |
| 9197173 | Denison et al. | Nov 2015 | B2 |
| 9198582 | Watson et al. | Dec 2015 | B2 |
| 9198586 | Melker | Dec 2015 | B2 |
| 9198604 | Venkatraman et al. | Dec 2015 | B2 |
| 9198616 | Addison et al. | Dec 2015 | B2 |
| 9198634 | Pretorius et al. | Dec 2015 | B2 |
| 9199078 | Gunderson | Dec 2015 | B1 |
| 9201902 | Doi et al. | Dec 2015 | B2 |
| 9202008 | Frederick et al. | Dec 2015 | B1 |
| 9204796 | Tran | Dec 2015 | B2 |
| 9208173 | Doi et al. | Dec 2015 | B1 |
| 9208557 | Pautot | Dec 2015 | B2 |
| 9208587 | Hardy et al. | Dec 2015 | B2 |
| 9209782 | Ricci et al. | Dec 2015 | B2 |
| 9211413 | Ziegler et al. | Dec 2015 | B2 |
| 9213990 | Adjaoute | Dec 2015 | B2 |
| 9215298 | Schiff | Dec 2015 | B2 |
| 9215980 | Tran et al. | Dec 2015 | B2 |
| 9215987 | Trayanova et al. | Dec 2015 | B2 |
| 9215991 | Inan et al. | Dec 2015 | B2 |
| 9216001 | Owen et al. | Dec 2015 | B2 |
| 9216065 | Cohen et al. | Dec 2015 | B2 |
| 9218181 | Chmiel et al. | Dec 2015 | B1 |
| 9220440 | Addison et al. | Dec 2015 | B2 |
| 9220455 | Sarrafzadeh et al. | Dec 2015 | B2 |
| 9220459 | Addison et al. | Dec 2015 | B2 |
| 9220460 | Addison et al. | Dec 2015 | B2 |
| 9220467 | Wu et al. | Dec 2015 | B2 |
| 9220856 | Martin et al. | Dec 2015 | B2 |
| 9223569 | Chmiel et al. | Dec 2015 | B1 |
| 9226660 | de Boer et al. | Jan 2016 | B2 |
| 9226665 | Tearney et al. | Jan 2016 | B2 |
| 9226676 | Chiu | Jan 2016 | B2 |
| 9236046 | Watson et al. | Jan 2016 | B2 |
| 9237855 | Hong et al. | Jan 2016 | B2 |
| 9239951 | Hoffberg et al. | Jan 2016 | B2 |
| 9241667 | Narayan et al. | Jan 2016 | B2 |
| 9242090 | Atalar et al. | Jan 2016 | B2 |
| 9242095 | Grill, Jr. et al. | Jan 2016 | B2 |
| 9245091 | Voigt et al. | Jan 2016 | B2 |
| 9247901 | Kamath et al. | Feb 2016 | B2 |
| 9248288 | Panken et al. | Feb 2016 | B2 |
| 9248306 | Joo et al. | Feb 2016 | B2 |
| 9249200 | Deisseroth et al. | Feb 2016 | B2 |
| 9254089 | Tearney et al. | Feb 2016 | B2 |
| 9254093 | Spector | Feb 2016 | B2 |
| 9254095 | Galloway et al. | Feb 2016 | B2 |
| 9254102 | Tearney et al. | Feb 2016 | B2 |
| 9254383 | Simon et al. | Feb 2016 | B2 |
| 9256701 | Chen et al. | Feb 2016 | B2 |
| 9258199 | Kalika et al. | Feb 2016 | B2 |
| 9258561 | Amon et al. | Feb 2016 | B2 |
| 9259167 | Lo et al. | Feb 2016 | B2 |
| 9259579 | Grill et al. | Feb 2016 | B2 |
| 9259591 | Brown et al. | Feb 2016 | B2 |
| 9261573 | Radparvar et al. | Feb 2016 | B1 |
| 9262826 | Khachaturian et al. | Feb 2016 | B2 |
| 9264877 | Feher | Feb 2016 | B2 |
| 9267936 | Hickman et al. | Feb 2016 | B2 |
| 9269127 | Ding et al. | Feb 2016 | B2 |
| RE45922 | Marcovecchio et al. | Mar 2016 | E |
| 9277871 | Keenan et al. | Mar 2016 | B2 |
| 9277956 | Zhang | Mar 2016 | B2 |
| 9278226 | Olson | Mar 2016 | B2 |
| 9282896 | Crawley et al. | Mar 2016 | B2 |
| 9282902 | Richards et al. | Mar 2016 | B2 |
| 9282908 | Spector | Mar 2016 | B2 |
| 9282910 | Narayan et al. | Mar 2016 | B2 |
| 9282911 | Stickney et al. | Mar 2016 | B2 |
| 9282925 | Kamath et al. | Mar 2016 | B2 |
| 9282931 | Tearney et al. | Mar 2016 | B2 |
| 9286662 | Wang et al. | Mar 2016 | B2 |
| 9287939 | Xu et al. | Mar 2016 | B2 |
| 9289133 | Cohen et al. | Mar 2016 | B2 |
| 9289136 | Addison et al. | Mar 2016 | B2 |
| 9289150 | Gupta et al. | Mar 2016 | B1 |
| 9289165 | Soykan et al. | Mar 2016 | B2 |
| 9289167 | Diab et al. | Mar 2016 | B2 |
| 9289471 | White | Mar 2016 | B2 |
| 9289545 | Olde et al. | Mar 2016 | B2 |
| 9294074 | Brockway | Mar 2016 | B2 |
| 9295391 | Tearney et al. | Mar 2016 | B1 |
| 9301703 | Sato | Apr 2016 | B2 |
| 9301705 | Halperin et al. | Apr 2016 | B2 |
| 9302111 | John et al. | Apr 2016 | B2 |
| 9304121 | Tearney et al. | Apr 2016 | B2 |
| 9305334 | Barzelay et al. | Apr 2016 | B2 |
| 9305350 | Crawley et al. | Apr 2016 | B2 |
| 9307407 | Feher | Apr 2016 | B1 |
| 9307917 | Hong et al. | Apr 2016 | B2 |
| 9308042 | Whayne et al. | Apr 2016 | B2 |
| 9308052 | Tolkowsky | Apr 2016 | B2 |
| 9314181 | Brockway et al. | Apr 2016 | B2 |
| 9314305 | Jenkins et al. | Apr 2016 | B2 |
| 9317626 | Chan | Apr 2016 | B2 |
| 9317740 | Ricanek, Jr. et al. | Apr 2016 | B2 |
| 9319028 | Ricci et al. | Apr 2016 | B2 |
| 9319212 | Feher | Apr 2016 | B2 |
| 9320491 | Konofagou et al. | Apr 2016 | B2 |
| 9321544 | Thompson et al. | Apr 2016 | B2 |
| 9323890 | Hayashi | Apr 2016 | B2 |
| 9324005 | Wadhwa et al. | Apr 2016 | B2 |
| 9324141 | Begin | Apr 2016 | B2 |
| 9324144 | Khachaturian et al. | Apr 2016 | B2 |
| 9325348 | Moroz et al. | Apr 2016 | B2 |
| 9326682 | Tearney et al. | May 2016 | B2 |
| 9326697 | Linker | May 2016 | B2 |
| 9326722 | Young | May 2016 | B2 |
| 9327130 | Marculescu et al. | May 2016 | B2 |
| 9330092 | Vakoc et al. | May 2016 | B2 |
| 9330459 | Crawley et al. | May 2016 | B2 |
| 9332357 | Adachi et al. | May 2016 | B2 |
| 9332939 | Osorio | May 2016 | B2 |
| 9332942 | Jaffer et al. | May 2016 | B2 |
| 9339202 | Brockway et al. | May 2016 | B2 |
| 9339206 | Grunwald | May 2016 | B2 |
| 9339230 | Kassab | May 2016 | B2 |
| 9339241 | Najarian et al. | May 2016 | B2 |
| 9339436 | Freeman et al. | May 2016 | B2 |
| 9339662 | Allavatam et al. | May 2016 | B2 |
| 9340589 | Deisseroth et al. | May 2016 | B2 |
| 9341783 | Namati et al. | May 2016 | B2 |
| 9345413 | Schie et al. | May 2016 | B2 |
| 9345609 | Hyde et al. | May 2016 | B2 |
| 9345888 | Wedan et al. | May 2016 | B2 |
| 9351640 | Tran | May 2016 | B2 |
| 9351642 | Nadkarni et al. | May 2016 | B2 |
| 9351649 | Mestha et al. | May 2016 | B2 |
| 9351661 | Kassab | May 2016 | B2 |
| 9351668 | Brauker et al. | May 2016 | B2 |
| 9351674 | Baker, Jr. | May 2016 | B2 |
| 9352057 | Ahrens et al. | May 2016 | B2 |
| 9352165 | Zhang | May 2016 | B2 |
| 9354115 | Darty | May 2016 | B2 |
| 9356731 | Baheti et al. | May 2016 | B2 |
| 20010028743 | Kostrzewski et al. | Oct 2001 | A1 |
| 20020009756 | Mandell et al. | Jan 2002 | A1 |
| 20020015532 | Kostrzewski et al. | Feb 2002 | A1 |
| 20020032386 | Sackner et al. | Mar 2002 | A1 |
| 20020040192 | Prutchi | Apr 2002 | A1 |
| 20020054694 | Vachtsevanos et al. | May 2002 | A1 |
| 20020059022 | Breed et al. | May 2002 | A1 |
| 20020082756 | Breed et al. | Jun 2002 | A1 |
| 20020123975 | Poluzzi et al. | Sep 2002 | A1 |
| 20020138014 | Baura | Sep 2002 | A1 |
| 20020151992 | Hoffberg et al. | Oct 2002 | A1 |
| 20020165854 | Blayvas et al. | Nov 2002 | A1 |
| 20020173936 | Srinivasan et al. | Nov 2002 | A1 |
| 20020176624 | Kostrzewski et al. | Nov 2002 | A1 |
| 20020186875 | Burmer et al. | Dec 2002 | A1 |
| 20020194159 | Kamath et al. | Dec 2002 | A1 |
| 20030036835 | Breed et al. | Feb 2003 | A1 |
| 20030045806 | Brydon | Mar 2003 | A1 |
| 20030055799 | Starzyk | Mar 2003 | A1 |
| 20030059837 | Levinson et al. | Mar 2003 | A1 |
| 20030061228 | Kamath et al. | Mar 2003 | A1 |
| 20030081836 | Averbuch et al. | May 2003 | A1 |
| 20030086593 | Liu et al. | May 2003 | A1 |
| 20030101164 | Pic et al. | May 2003 | A1 |
| 20030135097 | Wiederhold et al. | Jul 2003 | A1 |
| 20030135109 | Raghavan et al. | Jul 2003 | A1 |
| 20030135127 | Sackner et al. | Jul 2003 | A1 |
| 20030176656 | Mandell et al. | Sep 2003 | A1 |
| 20030200189 | Meng et al. | Oct 2003 | A1 |
| 20030208451 | Liaw | Nov 2003 | A1 |
| 20030209893 | Breed et al. | Nov 2003 | A1 |
| 20030217047 | Marchisio | Nov 2003 | A1 |
| 20030228565 | Oestreicher et al. | Dec 2003 | A1 |
| 20030233034 | Varri et al. | Dec 2003 | A1 |
| 20030233132 | Pastore et al. | Dec 2003 | A1 |
| 20040019470 | Card et al. | Jan 2004 | A1 |
| 20040045030 | Reynolds et al. | Mar 2004 | A1 |
| 20040068199 | Echauz et al. | Apr 2004 | A1 |
| 20040097802 | Cohen | May 2004 | A1 |
| 20040125121 | Pea et al. | Jul 2004 | A1 |
| 20040125133 | Pea et al. | Jul 2004 | A1 |
| 20040125148 | Pea et al. | Jul 2004 | A1 |
| 20040129478 | Breed et al. | Jul 2004 | A1 |
| 20040133355 | Schneider | Jul 2004 | A1 |
| 20040138578 | Pineda et al. | Jul 2004 | A1 |
| 20040207548 | Kilbank | Oct 2004 | A1 |
| 20040225629 | Eder | Nov 2004 | A1 |
| 20040225649 | Yeo et al. | Nov 2004 | A1 |
| 20040229210 | Sabry et al. | Nov 2004 | A1 |
| 20040230131 | Kassab et al. | Nov 2004 | A1 |
| 20040243567 | Levy | Dec 2004 | A1 |
| 20040267086 | Anstadt et al. | Dec 2004 | A1 |
| 20050008179 | Quinn | Jan 2005 | A1 |
| 20050017488 | Breed et al. | Jan 2005 | A1 |
| 20050021101 | Chen et al. | Jan 2005 | A1 |
| 20050026199 | Shaw | Feb 2005 | A1 |
| 20050027457 | Mandell et al. | Feb 2005 | A1 |
| 20050046584 | Breed | Mar 2005 | A1 |
| 20050069162 | Haykin et al. | Mar 2005 | A1 |
| 20050075846 | Kim | Apr 2005 | A1 |
| 20050076190 | Shaw | Apr 2005 | A1 |
| 20050079524 | Shaw | Apr 2005 | A1 |
| 20050089923 | Levinson et al. | Apr 2005 | A9 |
| 20050102246 | Movellan et al. | May 2005 | A1 |
| 20050113666 | Bonmassar et al. | May 2005 | A1 |
| 20050114078 | Srinivasan et al. | May 2005 | A1 |
| 20050119454 | Mandell et al. | Jun 2005 | A1 |
| 20050124864 | Mack et al. | Jun 2005 | A1 |
| 20050131607 | Breed | Jun 2005 | A1 |
| 20050131660 | Yadegar et al. | Jun 2005 | A1 |
| 20050137480 | Alt et al. | Jun 2005 | A1 |
| 20050144284 | Ludwig et al. | Jun 2005 | A1 |
| 20050156775 | Petre et al. | Jul 2005 | A1 |
| 20050158736 | Shaw | Jul 2005 | A1 |
| 20050183958 | Wikiel et al. | Aug 2005 | A1 |
| 20050196047 | Owechko et al. | Sep 2005 | A1 |
| 20050197590 | Osorio et al. | Sep 2005 | A1 |
| 20050203360 | Brauker et al. | Sep 2005 | A1 |
| 20050203434 | Kassab | Sep 2005 | A1 |
| 20050234289 | Anstadt et al. | Oct 2005 | A1 |
| 20050246314 | Eder | Nov 2005 | A1 |
| 20050248136 | Breed et al. | Nov 2005 | A1 |
| 20050267911 | Kamath et al. | Dec 2005 | A1 |
| 20050272110 | Drukier | Dec 2005 | A1 |
| 20050286179 | Huff et al. | Dec 2005 | A1 |
| 20060015492 | Keating et al. | Jan 2006 | A1 |
| 20060015494 | Keating et al. | Jan 2006 | A1 |
| 20060015495 | Keating et al. | Jan 2006 | A1 |
| 20060015496 | Keating et al. | Jan 2006 | A1 |
| 20060015497 | Keating et al. | Jan 2006 | A1 |
| 20060020597 | Keating et al. | Jan 2006 | A1 |
| 20060059028 | Eder | Mar 2006 | A1 |
| 20060084115 | DeNardo et al. | Apr 2006 | A1 |
| 20060094967 | Bennett et al. | May 2006 | A1 |
| 20060101017 | Eder | May 2006 | A1 |
| 20060106797 | Srinivasa et al. | May 2006 | A1 |
| 20060111754 | Rezai et al. | May 2006 | A1 |
| 20060120584 | Hillman | Jun 2006 | A1 |
| 20060123363 | Williams et al. | Jun 2006 | A1 |
| 20060142634 | Anstadt et al. | Jun 2006 | A1 |
| 20060149139 | Bonmassar et al. | Jul 2006 | A1 |
| 20060155398 | Hoffberg et al. | Jul 2006 | A1 |
| 20060167334 | Anstadt et al. | Jul 2006 | A1 |
| 20060167784 | Hoffberg | Jul 2006 | A1 |
| 20060184473 | Eder | Aug 2006 | A1 |
| 20060200253 | Hoffberg et al. | Sep 2006 | A1 |
| 20060200258 | Hoffberg et al. | Sep 2006 | A1 |
| 20060200259 | Hoffberg et al. | Sep 2006 | A1 |
| 20060200260 | Hoffberg et al. | Sep 2006 | A1 |
| 20060206159 | Moffitt et al. | Sep 2006 | A1 |
| 20060208169 | Breed et al. | Sep 2006 | A1 |
| 20060211909 | Anstadt et al. | Sep 2006 | A1 |
| 20060229822 | Theobald et al. | Oct 2006 | A1 |
| 20060241510 | Halperin et al. | Oct 2006 | A1 |
| 20060244246 | Breed et al. | Nov 2006 | A1 |
| 20060247536 | Koski et al. | Nov 2006 | A1 |
| 20060247542 | Watanabe et al. | Nov 2006 | A1 |
| 20060251293 | Piirainen et al. | Nov 2006 | A1 |
| 20060253258 | Miyake | Nov 2006 | A1 |
| 20060253781 | Pea et al. | Nov 2006 | A1 |
| 20060293607 | Alt et al. | Dec 2006 | A1 |
| 20070013509 | Lakshmanan et al. | Jan 2007 | A1 |
| 20070016476 | Hoffberg et al. | Jan 2007 | A1 |
| 20070025597 | Breed et al. | Feb 2007 | A1 |
| 20070035114 | Breed et al. | Feb 2007 | A1 |
| 20070053513 | Hoffberg | Mar 2007 | A1 |
| 20070054266 | Sato et al. | Mar 2007 | A1 |
| 20070054347 | Rosendahl et al. | Mar 2007 | A1 |
| 20070067003 | Sanchez et al. | Mar 2007 | A1 |
| 20070083128 | Cote et al. | Apr 2007 | A1 |
| 20070083243 | Prakash et al. | Apr 2007 | A1 |
| 20070086624 | Breed et al. | Apr 2007 | A1 |
| 20070087756 | Hoffberg | Apr 2007 | A1 |
| 20070103328 | Lakshmanan et al. | May 2007 | A1 |
| 20070106170 | Dunseath et al. | May 2007 | A1 |
| 20070118054 | Pinhas et al. | May 2007 | A1 |
| 20070135984 | Breed et al. | Jun 2007 | A1 |
| 20070140551 | He et al. | Jun 2007 | A1 |
| 20070150006 | Libbus et al. | Jun 2007 | A1 |
| 20070150021 | Chen et al. | Jun 2007 | A1 |
| 20070152433 | Weichenberger et al. | Jul 2007 | A1 |
| 20070154063 | Breed | Jul 2007 | A1 |
| 20070154078 | He et al. | Jul 2007 | A1 |
| 20070154079 | He et al. | Jul 2007 | A1 |
| 20070154099 | He et al. | Jul 2007 | A1 |
| 20070156317 | Breed | Jul 2007 | A1 |
| 20070160973 | Burns | Jul 2007 | A1 |
| 20070162084 | Chen et al. | Jul 2007 | A1 |
| 20070162189 | Huff et al. | Jul 2007 | A1 |
| 20070162992 | Burns | Jul 2007 | A1 |
| 20070167846 | Sternickel et al. | Jul 2007 | A1 |
| 20070175998 | Lev | Aug 2007 | A1 |
| 20070191742 | Park | Aug 2007 | A1 |
| 20070193811 | Breed et al. | Aug 2007 | A1 |
| 20070213786 | Sackellares et al. | Sep 2007 | A1 |
| 20070219749 | Jayabalan et al. | Sep 2007 | A1 |
| 20070230795 | Abramoff et al. | Oct 2007 | A1 |
| 20070256432 | Zugibe et al. | Nov 2007 | A1 |
| 20070258329 | Winey | Nov 2007 | A1 |
| 20070260285 | Libbus et al. | Nov 2007 | A1 |
| 20070260425 | Kim | Nov 2007 | A1 |
| 20070260427 | Kim | Nov 2007 | A1 |
| 20070260656 | Wiig | Nov 2007 | A1 |
| 20070262574 | Breed et al. | Nov 2007 | A1 |
| 20070265533 | Tran | Nov 2007 | A1 |
| 20070265806 | Kim | Nov 2007 | A1 |
| 20070265808 | Kim | Nov 2007 | A1 |
| 20070273504 | Tran | Nov 2007 | A1 |
| 20070276270 | Tran | Nov 2007 | A1 |
| 20070276279 | Echauz et al. | Nov 2007 | A1 |
| 20070286336 | Bernard et al. | Dec 2007 | A1 |
| 20070299349 | Alt et al. | Dec 2007 | A1 |
| 20080001735 | Tran | Jan 2008 | A1 |
| 20080004672 | Dalal et al. | Jan 2008 | A1 |
| 20080004904 | Tran | Jan 2008 | A1 |
| 20080013747 | Tran | Jan 2008 | A1 |
| 20080015871 | Eder | Jan 2008 | A1 |
| 20080021336 | Dobak | Jan 2008 | A1 |
| 20080021342 | Echauz et al. | Jan 2008 | A1 |
| 20080027769 | Eder | Jan 2008 | A1 |
| 20080027841 | Eder | Jan 2008 | A1 |
| 20080033316 | Kassab et al. | Feb 2008 | A1 |
| 20080036187 | Breed | Feb 2008 | A1 |
| 20080036580 | Breed | Feb 2008 | A1 |
| 20080039904 | Bulkes et al. | Feb 2008 | A1 |
| 20080042067 | Rousso et al. | Feb 2008 | A1 |
| 20080051660 | Kakadaris et al. | Feb 2008 | A1 |
| 20080058668 | Seyed Momen et al. | Mar 2008 | A1 |
| 20080065291 | Breed | Mar 2008 | A1 |
| 20080071136 | Oohashi et al. | Mar 2008 | A1 |
| 20080091114 | Min et al. | Apr 2008 | A1 |
| 20080114564 | Ihara | May 2008 | A1 |
| 20080119749 | Haro et al. | May 2008 | A1 |
| 20080128626 | Rousso et al. | Jun 2008 | A1 |
| 20080142713 | Breed et al. | Jun 2008 | A1 |
| 20080144944 | Breed | Jun 2008 | A1 |
| 20080146334 | Kil | Jun 2008 | A1 |
| 20080147438 | Kil | Jun 2008 | A1 |
| 20080147440 | Kil | Jun 2008 | A1 |
| 20080147441 | Kil | Jun 2008 | A1 |
| 20080152217 | Greer | Jun 2008 | A1 |
| 20080157510 | Breed et al. | Jul 2008 | A1 |
| 20080161877 | Kirby et al. | Jul 2008 | A1 |
| 20080162487 | Richter | Jul 2008 | A1 |
| 20080175480 | Shi et al. | Jul 2008 | A1 |
| 20080194946 | Summers et al. | Aug 2008 | A1 |
| 20080194975 | MacQuarrie et al. | Aug 2008 | A1 |
| 20080194996 | Kassab | Aug 2008 | A1 |
| 20080195261 | Breed | Aug 2008 | A1 |
| 20080222734 | Redlich et al. | Sep 2008 | A1 |
| 20080230702 | Rousso et al. | Sep 2008 | A1 |
| 20080230705 | Rousso et al. | Sep 2008 | A1 |
| 20080235165 | Movellan et al. | Sep 2008 | A1 |
| 20080236275 | Breed et al. | Oct 2008 | A1 |
| 20080247598 | Movellan et al. | Oct 2008 | A1 |
| 20080256069 | Eder | Oct 2008 | A1 |
| 20080262367 | Mugler et al. | Oct 2008 | A1 |
| 20080263323 | Mould et al. | Oct 2008 | A1 |
| 20080265130 | Colomb et al. | Oct 2008 | A1 |
| 20080269625 | Halperin et al. | Oct 2008 | A1 |
| 20080270328 | Lafferty et al. | Oct 2008 | A1 |
| 20080275314 | Mack et al. | Nov 2008 | A1 |
| 20080275349 | Halperin et al. | Nov 2008 | A1 |
| 20080292146 | Breed et al. | Nov 2008 | A1 |
| 20080294019 | Tran | Nov 2008 | A1 |
| 20080306397 | Bonmassar et al. | Dec 2008 | A1 |
| 20080312523 | Dunseath | Dec 2008 | A1 |
| 20090012766 | Miyake et al. | Jan 2009 | A1 |
| 20090018891 | Eder | Jan 2009 | A1 |
| 20090024044 | Virtanen et al. | Jan 2009 | A1 |
| 20090024549 | Johnson | Jan 2009 | A1 |
| 20090036758 | Brauker et al. | Feb 2009 | A1 |
| 20090036940 | Wei et al. | Feb 2009 | A1 |
| 20090043181 | Brauker et al. | Feb 2009 | A1 |
| 20090043182 | Brauker et al. | Feb 2009 | A1 |
| 20090043525 | Brauker et al. | Feb 2009 | A1 |
| 20090043541 | Brauker et al. | Feb 2009 | A1 |
| 20090043542 | Brauker et al. | Feb 2009 | A1 |
| 20090054742 | Kaminska et al. | Feb 2009 | A1 |
| 20090054758 | Dunseath | Feb 2009 | A1 |
| 20090055147 | Miyake et al. | Feb 2009 | A1 |
| 20090062635 | Brauker et al. | Mar 2009 | A1 |
| 20090062684 | Gregersen et al. | Mar 2009 | A1 |
| 20090076342 | Amurthur et al. | Mar 2009 | A1 |
| 20090076343 | James et al. | Mar 2009 | A1 |
| 20090076346 | James et al. | Mar 2009 | A1 |
| 20090076348 | Manicka et al. | Mar 2009 | A1 |
| 20090076349 | Libbus et al. | Mar 2009 | A1 |
| 20090076350 | Bly et al. | Mar 2009 | A1 |
| 20090076401 | Mazar et al. | Mar 2009 | A1 |
| 20090078875 | Rousso et al. | Mar 2009 | A1 |
| 20090083010 | Qi et al. | Mar 2009 | A1 |
| 20090099473 | Dunseath et al. | Apr 2009 | A1 |
| 20090105556 | Fricke et al. | Apr 2009 | A1 |
| 20090118637 | Kassab et al. | May 2009 | A1 |
| 20090169075 | Ishida et al. | Jul 2009 | A1 |
| 20090171740 | Eder | Jul 2009 | A1 |
| 20090177420 | Fournier et al. | Jul 2009 | A1 |
| 20090182287 | Kassab | Jul 2009 | A1 |
| 20090203972 | Heneghan et al. | Aug 2009 | A1 |
| 20090203981 | Brauker et al. | Aug 2009 | A1 |
| 20090204029 | Kassab | Aug 2009 | A1 |
| 20090204341 | Brauker et al. | Aug 2009 | A1 |
| 20090216133 | Kassab | Aug 2009 | A1 |
| 20090222108 | Lou et al. | Sep 2009 | A1 |
| 20090227876 | Tran | Sep 2009 | A1 |
| 20090227877 | Tran | Sep 2009 | A1 |
| 20090228408 | Kaushal et al. | Sep 2009 | A1 |
| 20090231173 | Kilbank | Sep 2009 | A1 |
| 20090240366 | Kaushal et al. | Sep 2009 | A1 |
| 20090259533 | Utter et al. | Oct 2009 | A1 |
| 20090259534 | Utter et al. | Oct 2009 | A1 |
| 20090259537 | Veksler et al. | Oct 2009 | A1 |
| 20090270746 | Min | Oct 2009 | A1 |
| 20090271342 | Eder | Oct 2009 | A1 |
| 20090285166 | Huber et al. | Nov 2009 | A1 |
| 20090286509 | Huber et al. | Nov 2009 | A1 |
| 20090286512 | Huber et al. | Nov 2009 | A1 |
| 20090288140 | Huber et al. | Nov 2009 | A1 |
| 20090288152 | Huber et al. | Nov 2009 | A1 |
| 20090299162 | Brauker et al. | Dec 2009 | A1 |
| 20090307164 | Baughman | Dec 2009 | A1 |
| 20090312612 | Rantala | Dec 2009 | A1 |
| 20090312819 | Klefenz et al. | Dec 2009 | A1 |
| 20090313041 | Eder | Dec 2009 | A1 |
| 20090316988 | Xu et al. | Dec 2009 | A1 |
| 20090318779 | Tran | Dec 2009 | A1 |
| 20090318987 | Kroll et al. | Dec 2009 | A1 |
| 20100010324 | Brauker et al. | Jan 2010 | A1 |
| 20100010331 | Brauker et al. | Jan 2010 | A1 |
| 20100010332 | Brauker et al. | Jan 2010 | A1 |
| 20100010355 | Kassab | Jan 2010 | A1 |
| 20100010368 | Kassab | Jan 2010 | A1 |
| 20100010488 | Kassab et al. | Jan 2010 | A1 |
| 20100010503 | Kassab | Jan 2010 | A1 |
| 20100010556 | Zhao et al. | Jan 2010 | A1 |
| 20100010681 | Zugibe et al. | Jan 2010 | A1 |
| 20100014718 | Savvides et al. | Jan 2010 | A1 |
| 20100016687 | Brauker et al. | Jan 2010 | A1 |
| 20100018718 | Krebs et al. | Jan 2010 | A1 |
| 20100020208 | Barbu | Jan 2010 | A1 |
| 20100020961 | Spottiswoode | Jan 2010 | A1 |
| 20100022855 | Brauker et al. | Jan 2010 | A1 |
| 20100023307 | Lee et al. | Jan 2010 | A1 |
| 20100026799 | Fujisawa et al. | Feb 2010 | A1 |
| 20100027431 | Morrison et al. | Feb 2010 | A1 |
| 20100027469 | Gurajala et al. | Feb 2010 | A1 |
| 20100027845 | Kim et al. | Feb 2010 | A1 |
| 20100027846 | Xu et al. | Feb 2010 | A1 |
| 20100027892 | Guan et al. | Feb 2010 | A1 |
| 20100030038 | Brauker et al. | Feb 2010 | A1 |
| 20100030090 | Zhang et al. | Feb 2010 | A1 |
| 20100030484 | Brauker et al. | Feb 2010 | A1 |
| 20100030485 | Brauker et al. | Feb 2010 | A1 |
| 20100041365 | Lott et al. | Feb 2010 | A1 |
| 20100045465 | Brauker et al. | Feb 2010 | A1 |
| 20100046842 | Conwell | Feb 2010 | A1 |
| 20100048242 | Rhoads et al. | Feb 2010 | A1 |
| 20100049369 | Lou et al. | Feb 2010 | A1 |
| 20100066540 | Theobald et al. | Mar 2010 | A1 |
| 20100069768 | Min et al. | Mar 2010 | A1 |
| 20100076642 | Hoffberg et al. | Mar 2010 | A1 |
| 20100094102 | Zhang et al. | Apr 2010 | A1 |
| 20100094147 | Inan et al. | Apr 2010 | A1 |
| 20100100148 | Min et al. | Apr 2010 | A1 |
| 20100100150 | Kirby et al. | Apr 2010 | A1 |
| 20100102825 | Bushnell et al. | Apr 2010 | A1 |
| 20100106269 | Garudadri et al. | Apr 2010 | A1 |
| 20100106458 | Leu et al. | Apr 2010 | A1 |
| 20100111396 | Boucheron | May 2010 | A1 |
| 20100113944 | Min et al. | May 2010 | A1 |
| 20100114207 | Snell et al. | May 2010 | A1 |
| 20100119128 | Zhang et al. | May 2010 | A1 |
| 20100121406 | Libbus et al. | May 2010 | A1 |
| 20100123587 | Walls | May 2010 | A1 |
| 20100130189 | Morrison et al. | May 2010 | A1 |
| 20100138026 | Kaushal et al. | Jun 2010 | A1 |
| 20100152795 | Schecter | Jun 2010 | A1 |
| 20100152905 | Kusiak | Jun 2010 | A1 |
| 20100159945 | Brisebois | Jun 2010 | A1 |
| 20100161654 | Levy | Jun 2010 | A1 |
| 20100168836 | Kassab | Jul 2010 | A1 |
| 20100169122 | Ruoff et al. | Jul 2010 | A1 |
| 20100169810 | Ruoff et al. | Jul 2010 | A1 |
| 20100174271 | Kassab | Jul 2010 | A1 |
| 20100179400 | Brauker et al. | Jul 2010 | A1 |
| 20100179438 | Heneghan et al. | Jul 2010 | A1 |
| 20100184702 | DeNardo et al. | Jul 2010 | A1 |
| 20100198098 | Osorio et al. | Aug 2010 | A1 |
| 20100204540 | Oohashi et al. | Aug 2010 | A1 |
| 20100204550 | Heneghan et al. | Aug 2010 | A1 |
| 20100210921 | Park et al. | Aug 2010 | A1 |
| 20100214545 | Funk et al. | Aug 2010 | A1 |
| 20100217145 | Buscema | Aug 2010 | A1 |
| 20100235285 | Hoffberg | Sep 2010 | A1 |
| 20100246544 | Brisebois et al. | Sep 2010 | A1 |
| 20100249628 | Kortelainen | Sep 2010 | A1 |
| 20100256701 | Muller | Oct 2010 | A1 |
| 20100272340 | Bar-Aviv et al. | Oct 2010 | A1 |
| 20100274219 | Wenzel et al. | Oct 2010 | A1 |
| 20100292968 | Gielis | Nov 2010 | A1 |
| 20100293115 | Seyed Momen | Nov 2010 | A1 |
| 20100304864 | Johnson et al. | Dec 2010 | A1 |
| 20100305634 | Moffitt et al. | Dec 2010 | A1 |
| 20100316283 | Greer | Dec 2010 | A1 |
| 20100317420 | Hoffberg | Dec 2010 | A1 |
| 20100331908 | Farazi | Dec 2010 | A1 |
| 20110004415 | Miyake et al. | Jan 2011 | A1 |
| 20110004513 | Hoffberg | Jan 2011 | A1 |
| 20110015468 | Aarts et al. | Jan 2011 | A1 |
| 20110015702 | Ternes et al. | Jan 2011 | A1 |
| 20110015703 | Ternes et al. | Jan 2011 | A1 |
| 20110015704 | Ternes et al. | Jan 2011 | A1 |
| 20110021928 | Giovangrandi et al. | Jan 2011 | A1 |
| 20110026832 | LeMoigne-Stewart et al. | Feb 2011 | A1 |
| 20110034811 | Naujokat et al. | Feb 2011 | A1 |
| 20110034824 | Kassab | Feb 2011 | A1 |
| 20110034967 | Chen et al. | Feb 2011 | A1 |
| 20110036801 | Krans et al. | Feb 2011 | A1 |
| 20110046498 | Klap et al. | Feb 2011 | A1 |
| 20110046508 | Wenzel et al. | Feb 2011 | A1 |
| 20110047105 | Sternickel et al. | Feb 2011 | A1 |
| 20110060230 | Gill et al. | Mar 2011 | A1 |
| 20110060235 | Crompvoets et al. | Mar 2011 | A1 |
| 20110066041 | Pandia | Mar 2011 | A1 |
| 20110066042 | Pandia et al. | Mar 2011 | A1 |
| 20110066205 | Crompvoets et al. | Mar 2011 | A1 |
| 20110066404 | Salazar-Tio et al. | Mar 2011 | A1 |
| 20110082511 | Aarts et al. | Apr 2011 | A1 |
| 20110087113 | Mack et al. | Apr 2011 | A1 |
| 20110087115 | Sackner et al. | Apr 2011 | A1 |
| 20110096144 | Pea et al. | Apr 2011 | A1 |
| 20110098583 | Pandia et al. | Apr 2011 | A1 |
| 20110098770 | Ryu et al. | Apr 2011 | A1 |
| 20110105930 | Thiagarajan et al. | May 2011 | A1 |
| 20110106558 | Solito et al. | May 2011 | A1 |
| 20110112426 | Causevic | May 2011 | A1 |
| 20110112442 | Meger et al. | May 2011 | A1 |
| 20110115624 | Tran | May 2011 | A1 |
| 20110118614 | Brauers et al. | May 2011 | A1 |
| 20110130670 | MacQuarrie et al. | Jun 2011 | A1 |
| 20110130671 | MacQuarrie et al. | Jun 2011 | A1 |
| 20110131041 | Cortez et al. | Jun 2011 | A1 |
| 20110131162 | Kaushal et al. | Jun 2011 | A1 |
| 20110135166 | Wechsler et al. | Jun 2011 | A1 |
| 20110137110 | Aarts et al. | Jun 2011 | A1 |
| 20110144065 | Denardo et al. | Jun 2011 | A1 |
| 20110144519 | Causevic | Jun 2011 | A1 |
| 20110150253 | Corona-Strauss et al. | Jun 2011 | A1 |
| 20110152974 | Rezai et al. | Jun 2011 | A1 |
| 20110156896 | Hoffberg et al. | Jun 2011 | A1 |
| 20110160656 | Johnson et al. | Jun 2011 | A1 |
| 20110164783 | Hays et al. | Jul 2011 | A1 |
| 20110167110 | Hoffberg et al. | Jul 2011 | A1 |
| 20110172500 | Van Dooren et al. | Jul 2011 | A1 |
| 20110172930 | Pancoska et al. | Jul 2011 | A1 |
| 20110181422 | Tran | Jul 2011 | A1 |
| 20110196254 | Wenzel et al. | Aug 2011 | A1 |
| 20110196441 | Ryu et al. | Aug 2011 | A1 |
| 20110196442 | Ryu et al. | Aug 2011 | A1 |
| 20110208016 | Bombardini | Aug 2011 | A1 |
| 20110236922 | Burns | Sep 2011 | A1 |
| 20110251502 | Friedrich et al. | Oct 2011 | A1 |
| 20110261178 | Lo et al. | Oct 2011 | A1 |
| 20110263958 | Brauker et al. | Oct 2011 | A1 |
| 20110263994 | Burns et al. | Oct 2011 | A1 |
| 20110268328 | Bar-Aviv et al. | Nov 2011 | A1 |
| 20110275364 | Austin et al. | Nov 2011 | A1 |
| 20110280447 | Conwell | Nov 2011 | A1 |
| 20110285982 | Breed | Nov 2011 | A1 |
| 20110301660 | Libbus et al. | Dec 2011 | A1 |
| 20110313285 | Fallavollita et al. | Dec 2011 | A1 |
| 20110319776 | Sweeney et al. | Dec 2011 | A1 |
| 20110319778 | Sweeney et al. | Dec 2011 | A1 |
| 20110319782 | Sweeney et al. | Dec 2011 | A1 |
| 20120004564 | Dobak | Jan 2012 | A1 |
| 20120004854 | Fernandez et al. | Jan 2012 | A1 |
| 20120010677 | Wenzel et al. | Jan 2012 | A1 |
| 20120010867 | Eder | Jan 2012 | A1 |
| 20120022336 | Teixeira | Jan 2012 | A1 |
| 20120022350 | Teixeira | Jan 2012 | A1 |
| 20120022384 | Teixeira | Jan 2012 | A1 |
| 20120022844 | Teixeira | Jan 2012 | A1 |
| 20120036016 | Hoffberg et al. | Feb 2012 | A1 |
| 20120041330 | Prichep et al. | Feb 2012 | A1 |
| 20120041608 | Zugibe et al. | Feb 2012 | A1 |
| 20120050750 | Hays et al. | Mar 2012 | A1 |
| 20120053441 | Kassab | Mar 2012 | A1 |
| 20120066217 | Eder | Mar 2012 | A1 |
| 20120066259 | Huber et al. | Mar 2012 | A1 |
| 20120071792 | Pfeffer et al. | Mar 2012 | A1 |
| 20120073825 | Routh et al. | Mar 2012 | A1 |
| 20120083246 | Huber et al. | Apr 2012 | A1 |
| 20120089046 | Kassab et al. | Apr 2012 | A1 |
| 20120092156 | Tran | Apr 2012 | A1 |
| 20120092157 | Tran | Apr 2012 | A1 |
| 20120095352 | Tran | Apr 2012 | A1 |
| 20120095357 | Tran | Apr 2012 | A1 |
| 20120109612 | Krebs et al. | May 2012 | A1 |
| 20120109653 | Yen et al. | May 2012 | A1 |
| 20120114249 | Conwell | May 2012 | A1 |
| 20120123232 | Najarian et al. | May 2012 | A1 |
| 20120123279 | Brueser et al. | May 2012 | A1 |
| 20120132211 | Halperin et al. | May 2012 | A1 |
| 20120136263 | Prakash et al. | May 2012 | A1 |
| 20120143072 | Prakash et al. | Jun 2012 | A1 |
| 20120143078 | Kassab et al. | Jun 2012 | A1 |
| 20120143382 | Lou | Jun 2012 | A1 |
| 20120148149 | Kumar et al. | Jun 2012 | A1 |
| 20120148157 | Kumar et al. | Jun 2012 | A1 |
| 20120157798 | Averina et al. | Jun 2012 | A1 |
| 20120157856 | An et al. | Jun 2012 | A1 |
| 20120157861 | Jarverud et al. | Jun 2012 | A1 |
| 20120158633 | Eder | Jun 2012 | A1 |
| 20120165892 | Min et al. | Jun 2012 | A1 |
| 20120169053 | Tchoryk, Jr. et al. | Jul 2012 | A1 |
| 20120172741 | Arcot-Krishnamurthy et al. | Jul 2012 | A1 |
| 20120172742 | Arcot-Krishnamurthy et al. | Jul 2012 | A1 |
| 20120172743 | Aguilar et al. | Jul 2012 | A1 |
| 20120172746 | Kassab | Jul 2012 | A1 |
| 20120173154 | Rosendahl et al. | Jul 2012 | A1 |
| 20120179216 | Moffitt et al. | Jul 2012 | A1 |
| 20120185012 | Ryu et al. | Jul 2012 | A1 |
| 20120190404 | Rhoads | Jul 2012 | A1 |
| 20120197333 | Zhao et al. | Aug 2012 | A1 |
| 20120197831 | Dong et al. | Aug 2012 | A1 |
| 20120203077 | He et al. | Aug 2012 | A1 |
| 20120203090 | Min | Aug 2012 | A1 |
| 20120209798 | Kaushal et al. | Aug 2012 | A1 |
| 20120214510 | Brisebois | Aug 2012 | A1 |
| 20120220835 | Chung | Aug 2012 | A1 |
| 20120226126 | Busse et al. | Sep 2012 | A1 |
| 20120238800 | Naujokat et al. | Sep 2012 | A1 |
| 20120242501 | Tran et al. | Sep 2012 | A1 |
| 20120245464 | Tran | Sep 2012 | A1 |
| 20120245476 | Skerl et al. | Sep 2012 | A1 |
| 20120245481 | Blanco et al. | Sep 2012 | A1 |
| 20120253419 | Rosenberg et al. | Oct 2012 | A1 |
| 20120257046 | Mueller et al. | Oct 2012 | A1 |
| 20120265350 | Ashdown | Oct 2012 | A1 |
| 20120271177 | Emerson et al. | Oct 2012 | A1 |
| 20120271371 | Keel et al. | Oct 2012 | A1 |
| 20120271382 | Arcot-Krishnamurthy et al. | Oct 2012 | A1 |
| 20120274937 | Hays et al. | Nov 2012 | A1 |
| 20120277545 | Teixeira | Nov 2012 | A1 |
| 20120290505 | Eder | Nov 2012 | A1 |
| 20120296228 | Zhang et al. | Nov 2012 | A1 |
| 20120303504 | Eder | Nov 2012 | A1 |
| 20120303560 | Sedaghat et al. | Nov 2012 | A1 |
| 20120330109 | Tran | Dec 2012 | A1 |
| 20120330170 | Chiu et al. | Dec 2012 | A1 |
| 20120330373 | Ternes et al. | Dec 2012 | A1 |
| 20130006317 | Keel et al. | Jan 2013 | A1 |
| 20130009783 | Tran | Jan 2013 | A1 |
| 20130011062 | Conwell et al. | Jan 2013 | A1 |
| 20130023956 | Ternes et al. | Jan 2013 | A1 |
| 20130023957 | Ternes et al. | Jan 2013 | A1 |
| 20130028052 | Routh et al. | Jan 2013 | A1 |
| 20130030312 | Keel et al. | Jan 2013 | A1 |
| 20130030314 | Keel et al. | Jan 2013 | A1 |
| 20130030315 | Keel et al. | Jan 2013 | A1 |
| 20130030484 | Zhang et al. | Jan 2013 | A1 |
| 20130030486 | Betzold | Jan 2013 | A1 |
| 20130030487 | Keel et al. | Jan 2013 | A1 |
| 20130053907 | Kirchner et al. | Feb 2013 | A1 |
| 20130053912 | Bornzin et al. | Feb 2013 | A1 |
| 20130053913 | Koh et al. | Feb 2013 | A1 |
| 20130053926 | Hincapie Ordonez et al. | Feb 2013 | A1 |
| 20130060296 | Wenzel et al. | Mar 2013 | A1 |
| 20130060297 | Stein | Mar 2013 | A1 |
| 20130063613 | Conwell | Mar 2013 | A1 |
| 20130069780 | Tran et al. | Mar 2013 | A1 |
| 20130071837 | Winters-Hilt et al. | Mar 2013 | A1 |
| 20130072807 | Tran | Mar 2013 | A1 |
| 20130073490 | Baughman et al. | Mar 2013 | A1 |
| 20130073981 | Pea et al. | Mar 2013 | A1 |
| 20130077843 | Bruder et al. | Mar 2013 | A1 |
| 20130077891 | Nimnual et al. | Mar 2013 | A1 |
| 20130079002 | Huber et al. | Mar 2013 | A1 |
| 20130085401 | Zhang et al. | Apr 2013 | A1 |
| 20130090247 | Gardner | Apr 2013 | A1 |
| 20130090265 | Gardner | Apr 2013 | A1 |
| 20130090266 | Gardner | Apr 2013 | A1 |
| 20130095459 | Tran | Apr 2013 | A1 |
| 20130096394 | Gupta et al. | Apr 2013 | A1 |
| 20130109989 | Busse et al. | May 2013 | A1 |
| 20130109995 | Rothman et al. | May 2013 | A1 |
| 20130123666 | Giuffrida et al. | May 2013 | A1 |
| 20130123684 | Giuffrida et al. | May 2013 | A1 |
| 20130123873 | Libbus et al. | May 2013 | A1 |
| 20130135008 | Zhang et al. | May 2013 | A1 |
| 20130137998 | Lange et al. | May 2013 | A1 |
| 20130144178 | Halperin et al. | Jun 2013 | A1 |
| 20130151447 | Kaushal et al. | Jun 2013 | A1 |
| 20130158415 | Kim et al. | Jun 2013 | A1 |
| 20130165819 | Tieu | Jun 2013 | A1 |
| 20130171599 | Bleich et al. | Jul 2013 | A1 |
| 20130172691 | Tran | Jul 2013 | A1 |
| 20130173194 | Dholakia et al. | Jul 2013 | A1 |
| 20130178718 | Tran et al. | Jul 2013 | A1 |
| 20130178730 | Sternickel et al. | Jul 2013 | A1 |
| 20130184538 | Lee et al. | Jul 2013 | A1 |
| 20130184553 | Kassab et al. | Jul 2013 | A1 |
| 20130184603 | Rothman | Jul 2013 | A1 |
| 20130189977 | Brisebois et al. | Jul 2013 | A1 |
| 20130190645 | Wenzel et al. | Jul 2013 | A1 |
| 20130190835 | Zhao et al. | Jul 2013 | A1 |
| 20130191090 | Krebs et al. | Jul 2013 | A1 |
| 20130197322 | Tran | Aug 2013 | A1 |
| 20130197375 | Heise et al. | Aug 2013 | A1 |
| 20130197380 | Oral et al. | Aug 2013 | A1 |
| 20130197597 | Anderson et al. | Aug 2013 | A1 |
| 20130202177 | Bar-Aviv et al. | Aug 2013 | A1 |
| 20130204122 | Hendler et al. | Aug 2013 | A1 |
| 20130211271 | Kang et al. | Aug 2013 | A1 |
| 20130211291 | Tran | Aug 2013 | A1 |
| 20130211482 | Piipponen | Aug 2013 | A1 |
| 20130214943 | Yen et al. | Aug 2013 | A1 |
| 20130218156 | Kassab et al. | Aug 2013 | A1 |
| 20130231574 | Tran | Sep 2013 | A1 |
| 20130236067 | Savvides et al. | Sep 2013 | A1 |
| 20130245502 | Lange et al. | Sep 2013 | A1 |
| 20130245722 | Ternes et al. | Sep 2013 | A1 |
| 20130252604 | Huber et al. | Sep 2013 | A1 |
| 20130261473 | Xi et al. | Oct 2013 | A1 |
| 20130269376 | Zugibe et al. | Oct 2013 | A1 |
| 20130273968 | Rhoads et al. | Oct 2013 | A1 |
| 20130296962 | Keel et al. | Nov 2013 | A1 |
| 20130301889 | Abramoff et al. | Nov 2013 | A1 |
| 20130303119 | Huber et al. | Nov 2013 | A1 |
| 20130303941 | Porges et al. | Nov 2013 | A1 |
| 20130310700 | Wiard et al. | Nov 2013 | A1 |
| 20130314694 | Tchoryk, Jr. et al. | Nov 2013 | A1 |
| 20130317392 | Kassab | Nov 2013 | A1 |
| 20130331661 | Woodward | Dec 2013 | A1 |
| 20130331904 | Ternes et al. | Dec 2013 | A1 |
| 20130336594 | Dorairaj | Dec 2013 | A1 |
| 20130338460 | He et al. | Dec 2013 | A1 |
| 20130338468 | Kassab | Dec 2013 | A1 |
| 20130338496 | Hielscher et al. | Dec 2013 | A1 |
| 20130338530 | Kassab | Dec 2013 | A1 |
| 20130345591 | Hincapie Ordonez et al. | Dec 2013 | A1 |
| 20140005496 | Sison et al. | Jan 2014 | A1 |
| 20140005743 | Giuffrida et al. | Jan 2014 | A1 |
| 20140012099 | Halperin et al. | Jan 2014 | A1 |
| 20140012144 | Crone | Jan 2014 | A1 |
| 20140025304 | McLennan et al. | Jan 2014 | A1 |
| 20140039330 | Seo et al. | Feb 2014 | A1 |
| 20140039333 | Min | Feb 2014 | A1 |
| 20140046209 | Klap et al. | Feb 2014 | A1 |
| 20140052209 | Ternes et al. | Feb 2014 | A1 |
| 20140052379 | McVay et al. | Feb 2014 | A1 |
| 20140055284 | Tran et al. | Feb 2014 | A1 |
| 20140066738 | Kassab | Mar 2014 | A1 |
| 20140066798 | Albert | Mar 2014 | A1 |
| 20140067463 | Richter | Mar 2014 | A1 |
| 20140067470 | Richter | Mar 2014 | A1 |
| 20140067484 | Richter | Mar 2014 | A1 |
| 20140067485 | Richter | Mar 2014 | A1 |
| 20140074179 | Heldman et al. | Mar 2014 | A1 |
| 20140074180 | Heldman et al. | Mar 2014 | A1 |
| 20140074564 | Richter | Mar 2014 | A1 |
| 20140077946 | Tran | Mar 2014 | A1 |
| 20140079297 | Tadayon et al. | Mar 2014 | A1 |
| 20140088415 | Hielscher et al. | Mar 2014 | A1 |
| 20140088676 | Moffitt et al. | Mar 2014 | A1 |
| 20140089241 | Hoffberg et al. | Mar 2014 | A1 |
| 20140094875 | Ternes et al. | Apr 2014 | A1 |
| 20140104059 | Tran | Apr 2014 | A1 |
| 20140114165 | Walker et al. | Apr 2014 | A1 |
| 20140114370 | Libbus et al. | Apr 2014 | A1 |
| 20140121476 | Tran et al. | May 2014 | A1 |
| 20140127672 | Davis et al. | May 2014 | A1 |
| 20140128953 | Zhao et al. | May 2014 | A1 |
| 20140135634 | Pranevicius et al. | May 2014 | A1 |
| 20140135645 | Wenzel et al. | May 2014 | A9 |
| 20140142437 | Inan et al. | May 2014 | A1 |
| 20140142444 | Ngo et al. | May 2014 | A1 |
| 20140142451 | Kim et al. | May 2014 | A1 |
| 20140143064 | Tran | May 2014 | A1 |
| 20140143251 | Wang et al. | May 2014 | A1 |
| 20140151563 | Rousso et al. | Jun 2014 | A1 |
| 20140163343 | Heneghan et al. | Jun 2014 | A1 |
| 20140163368 | Rousso et al. | Jun 2014 | A1 |
| 20140163425 | Tran | Jun 2014 | A1 |
| 20140169686 | Conwell et al. | Jun 2014 | A1 |
| 20140173452 | Hoffberg et al. | Jun 2014 | A1 |
| 20140180049 | Brauker et al. | Jun 2014 | A1 |
| 20140193087 | Conwell | Jul 2014 | A1 |
| 20140194702 | Tran | Jul 2014 | A1 |
| 20140200428 | Kassab | Jul 2014 | A1 |
| 20140200429 | Spector et al. | Jul 2014 | A1 |
| 20140200430 | Spector | Jul 2014 | A1 |
| 20140200471 | Spector | Jul 2014 | A1 |
| 20140200571 | Spector | Jul 2014 | A1 |
| 20140200572 | Spector | Jul 2014 | A1 |
| 20140200575 | Spector | Jul 2014 | A1 |
| 20140201126 | Zadeh et al. | Jul 2014 | A1 |
| 20140204700 | Valero et al. | Jul 2014 | A1 |
| 20140207204 | Halperin et al. | Jul 2014 | A1 |
| 20140213909 | Mestha et al. | Jul 2014 | A1 |
| 20140219566 | Rodriguez et al. | Aug 2014 | A1 |
| 20140221786 | Zhang et al. | Aug 2014 | A1 |
| 20140221859 | Albert | Aug 2014 | A1 |
| 20140222115 | Ternes et al. | Aug 2014 | A1 |
| 20140229409 | Kaushal et al. | Aug 2014 | A1 |
| 20140235201 | Gurajala et al. | Aug 2014 | A1 |
| 20140235965 | Tran | Aug 2014 | A1 |
| 20140236530 | Greene et al. | Aug 2014 | A1 |
| 20140249429 | Tran | Sep 2014 | A1 |
| 20140266787 | Tran | Sep 2014 | A1 |
| 20140267299 | Couse | Sep 2014 | A1 |
| 20140275824 | Couse | Sep 2014 | A1 |
| 20140275829 | Berezhnyy et al. | Sep 2014 | A1 |
| 20140275886 | Teixeira | Sep 2014 | A1 |
| 20140275925 | Thakur et al. | Sep 2014 | A1 |
| 20140276121 | Kassab | Sep 2014 | A1 |
| 20140276191 | Kassab | Sep 2014 | A1 |
| 20140277239 | Maskara et al. | Sep 2014 | A1 |
| 20140277241 | Bleich et al. | Sep 2014 | A1 |
| 20140288442 | Bombardini | Sep 2014 | A1 |
| 20140288551 | Bharmi et al. | Sep 2014 | A1 |
| 20140289172 | Rothman et al. | Sep 2014 | A1 |
| 20140308930 | Tran | Oct 2014 | A1 |
| 20140309543 | Keel et al. | Oct 2014 | A1 |
| 20140309959 | Shen et al. | Oct 2014 | A1 |
| 20140315576 | Brisebois | Oct 2014 | A1 |
| 20140316221 | Rothman | Oct 2014 | A1 |
| 20140321756 | Guan et al. | Oct 2014 | A9 |
| 20140323821 | Manicka et al. | Oct 2014 | A1 |
| 20140325019 | Austin et al. | Oct 2014 | A1 |
| 20140342703 | Huber et al. | Nov 2014 | A1 |
| 20140343396 | Sternickel et al. | Nov 2014 | A1 |
| 20140344013 | Karty et al. | Nov 2014 | A1 |
| 20140350361 | De Chazal et al. | Nov 2014 | A1 |
| 20140351188 | Bagg | Nov 2014 | A1 |
| 20140362013 | Nikoozadeh et al. | Dec 2014 | A1 |
| 20140364721 | Lee et al. | Dec 2014 | A1 |
| 20140371574 | Shusterman et al. | Dec 2014 | A1 |
| 20140371834 | Chen et al. | Dec 2014 | A1 |
| 20140378849 | Krimsky et al. | Dec 2014 | A1 |
| 20150005594 | Chamoun et al. | Jan 2015 | A1 |
| 20150005646 | Balakrishnan et al. | Jan 2015 | A1 |
| 20150005655 | Sato | Jan 2015 | A1 |
| 20150012256 | Routh et al. | Jan 2015 | A1 |
| 20150016702 | Huizenga et al. | Jan 2015 | A1 |
| 20150018632 | Khair | Jan 2015 | A1 |
| 20150018637 | Chen et al. | Jan 2015 | A1 |
| 20150025328 | Khair | Jan 2015 | A1 |
| 20150025334 | Jain | Jan 2015 | A1 |
| 20150025335 | Jain et al. | Jan 2015 | A1 |
| 20150025336 | Mestek et al. | Jan 2015 | A1 |
| 20150025393 | Hong et al. | Jan 2015 | A1 |
| 20150025394 | Hong et al. | Jan 2015 | A1 |
| 20150031964 | Bly et al. | Jan 2015 | A1 |
| 20150031969 | Khair | Jan 2015 | A1 |
| 20150032015 | Chiu | Jan 2015 | A1 |
| 20150032178 | Simon et al. | Jan 2015 | A1 |
| 20150038856 | Houlton et al. | Feb 2015 | A1 |
| 20150045684 | Schie et al. | Feb 2015 | A1 |
| 20150046095 | Takahashi et al. | Feb 2015 | A1 |
| 20150046181 | Adjaoute | Feb 2015 | A1 |
| 20150046216 | Adjaoute | Feb 2015 | A1 |
| 20150051083 | Regensburger et al. | Feb 2015 | A1 |
| 20150051452 | Ciaccio | Feb 2015 | A1 |
| 20150057512 | Kapoor | Feb 2015 | A1 |
| 20150058061 | Salama et al. | Feb 2015 | A1 |
| 20150065814 | Kapoor | Mar 2015 | A1 |
| 20150065815 | Najarian et al. | Mar 2015 | A1 |
| 20150065835 | Kassab | Mar 2015 | A1 |
| 20150068069 | Tran et al. | Mar 2015 | A1 |
| 20150073234 | Inan et al. | Mar 2015 | A1 |
| 20150073495 | Grill, Jr. et al. | Mar 2015 | A1 |
| 20150080746 | Bleich et al. | Mar 2015 | A1 |
| 20150081077 | Li et al. | Mar 2015 | A1 |
| 20150081324 | Adjaoute | Mar 2015 | A1 |
| 20150081911 | Li et al. | Mar 2015 | A1 |
| 20150086087 | Ricanek, Jr. et al. | Mar 2015 | A1 |
| 20150086131 | Wang et al. | Mar 2015 | A1 |
| 20150087589 | Schiemann et al. | Mar 2015 | A1 |
| 20150087931 | Banerjee et al. | Mar 2015 | A1 |
| 20150087947 | Freeman et al. | Mar 2015 | A1 |
| 20150088004 | Berckmans et al. | Mar 2015 | A1 |
| 20150088016 | Fleischacker et al. | Mar 2015 | A1 |
| 20150088024 | Sackellares et al. | Mar 2015 | A1 |
| 20150088214 | Allavatam et al. | Mar 2015 | A1 |
| 20150093037 | Dorairaj | Apr 2015 | A1 |
| 20150094012 | Gurajala et al. | Apr 2015 | A1 |
| 20150095146 | Adjaoute | Apr 2015 | A1 |
| 20150099941 | Tran | Apr 2015 | A1 |
| 20150099946 | Sahin | Apr 2015 | A1 |
| 20150103360 | Addison et al. | Apr 2015 | A1 |
| 20150103681 | Thiel et al. | Apr 2015 | A1 |
| 20150105086 | Thiel et al. | Apr 2015 | A1 |
| 20150105631 | Tran et al. | Apr 2015 | A1 |
| 20150105681 | Bonan et al. | Apr 2015 | A1 |
| 20150105695 | Sabatino | Apr 2015 | A1 |
| 20150106020 | Chung et al. | Apr 2015 | A1 |
| 20150106069 | Gielis | Apr 2015 | A1 |
| 20150106310 | Birdwell et al. | Apr 2015 | A1 |
| 20150106311 | Birdwell et al. | Apr 2015 | A1 |
| 20150106314 | Birdwell et al. | Apr 2015 | A1 |
| 20150106315 | Birdwell et al. | Apr 2015 | A1 |
| 20150106316 | Birdwell et al. | Apr 2015 | A1 |
| 20150109124 | He et al. | Apr 2015 | A1 |
| 20150112154 | He et al. | Apr 2015 | A1 |
| 20150112155 | Bijjani et al. | Apr 2015 | A1 |
| 20150112156 | He et al. | Apr 2015 | A1 |
| 20150112157 | Bijjani et al. | Apr 2015 | A1 |
| 20150112158 | He et al. | Apr 2015 | A1 |
| 20150112159 | He et al. | Apr 2015 | A1 |
| 20150112208 | He et al. | Apr 2015 | A1 |
| 20150112209 | Blaber et al. | Apr 2015 | A1 |
| 20150112211 | Purdy | Apr 2015 | A1 |
| 20150112212 | Purdy | Apr 2015 | A1 |
| 20150112220 | Sana et al. | Apr 2015 | A1 |
| 20150112403 | Ruffini et al. | Apr 2015 | A1 |
| 20150112409 | Hagedorn | Apr 2015 | A1 |
| 20150112452 | He et al. | Apr 2015 | A1 |
| 20150112606 | He et al. | Apr 2015 | A1 |
| 20150112636 | Grichnik et al. | Apr 2015 | A1 |
| 20150117741 | Imamura | Apr 2015 | A1 |
| 20150118158 | Jensen et al. | Apr 2015 | A1 |
| 20150119711 | Osumi et al. | Apr 2015 | A1 |
| 20150122018 | Yuen | May 2015 | A1 |
| 20150125832 | Tran | May 2015 | A1 |
| 20150126833 | Anderson et al. | May 2015 | A1 |
| 20150126848 | Baker et al. | May 2015 | A1 |
| 20150126876 | Lee et al. | May 2015 | A1 |
| 20150127061 | Ternes et al. | May 2015 | A1 |
| 20150127066 | Grill, Jr. et al. | May 2015 | A1 |
| 20150127067 | Ternes et al. | May 2015 | A1 |
| 20150133306 | Cronin | May 2015 | A1 |
| 20150133307 | Zhang et al. | May 2015 | A1 |
| 20150133795 | Tomaselli et al. | May 2015 | A1 |
| 20150133807 | Moorman et al. | May 2015 | A1 |
| 20150134019 | Moffitt et al. | May 2015 | A1 |
| 20150134315 | Sarmiento et al. | May 2015 | A1 |
| 20150135310 | Lee | May 2015 | A1 |
| 20150137988 | Gravenstein et al. | May 2015 | A1 |
| 20150139977 | Weiner et al. | May 2015 | A1 |
| 20150141846 | Sanchez et al. | May 2015 | A1 |
| 20150141857 | Nallathambi et al. | May 2015 | A1 |
| 20150141860 | Linker | May 2015 | A1 |
| 20150141861 | Trayanova et al. | May 2015 | A1 |
| 20150141863 | Kassab et al. | May 2015 | A1 |
| 20150142069 | Sambelashvili | May 2015 | A1 |
| 20150142070 | Sambelashvili | May 2015 | A1 |
| 20150148635 | Benaron | May 2015 | A1 |
| 20150157218 | Ahmad et al. | Jun 2015 | A1 |
| 20150157239 | Rissacher et al. | Jun 2015 | A1 |
| 20150157258 | Beattie et al. | Jun 2015 | A1 |
| 20150157269 | Lisogurski et al. | Jun 2015 | A1 |
| 20150157387 | OuYang et al. | Jun 2015 | A1 |
| 20150161629 | Verma | Jun 2015 | A1 |
| 20150164339 | Xu et al. | Jun 2015 | A1 |
| 20150164340 | Bedingham et al. | Jun 2015 | A1 |
| 20150164349 | Gopalakrishnan et al. | Jun 2015 | A1 |
| 20150164355 | Brockway et al. | Jun 2015 | A1 |
| 20150164358 | Moorman et al. | Jun 2015 | A1 |
| 20150164375 | Schindhelm et al. | Jun 2015 | A1 |
| 20150164433 | Halperin et al. | Jun 2015 | A1 |
| 20150165223 | Babaeizadeh et al. | Jun 2015 | A1 |
| 20150170052 | Abdul Rahman et al. | Jun 2015 | A1 |
| 20150171998 | Baheti et al. | Jun 2015 | A1 |
| 20150173631 | Richards et al. | Jun 2015 | A1 |
| 20150174307 | Eckman et al. | Jun 2015 | A1 |
| 20150174408 | Grill, Jr. et al. | Jun 2015 | A1 |
| 20150178631 | Thomas et al. | Jun 2015 | A1 |
| 20150181822 | Daetwyler et al. | Jul 2015 | A1 |
| 20150182160 | Kim et al. | Jul 2015 | A1 |
| 20150190060 | Addison et al. | Jul 2015 | A1 |
| 20150190088 | Chen et al. | Jul 2015 | A1 |
| 20150190636 | Simon et al. | Jul 2015 | A1 |
| 20150190637 | Simon et al. | Jul 2015 | A1 |
| 20150196213 | Pandia et al. | Jul 2015 | A1 |
| 20150196256 | Venkatraman et al. | Jul 2015 | A1 |
| 20150201853 | Hong et al. | Jul 2015 | A1 |
| 20150201854 | Hong et al. | Jul 2015 | A1 |
| 20150201859 | Baker et al. | Jul 2015 | A1 |
| 20150204559 | Hoffberg et al. | Jul 2015 | A1 |
| 20150206214 | Adjaoute | Jul 2015 | A1 |
| 20150208939 | Zigel et al. | Jul 2015 | A1 |
| 20150216426 | Burton et al. | Aug 2015 | A1 |
| 20150216431 | Pahlevan et al. | Aug 2015 | A1 |
| 20150216433 | Thakur et al. | Aug 2015 | A1 |
| 20150216480 | Nagata et al. | Aug 2015 | A1 |
| 20150216483 | Sevcencu et al. | Aug 2015 | A1 |
| 20150220486 | Karakonstantis et al. | Aug 2015 | A1 |
| 20150223708 | Richards et al. | Aug 2015 | A1 |
| 20150223711 | Raeder et al. | Aug 2015 | A1 |
| 20150223733 | Al-Alusi | Aug 2015 | A1 |
| 20150223756 | Nagata et al. | Aug 2015 | A1 |
| 20150223863 | Ghosh | Aug 2015 | A1 |
| 20150234976 | Chen et al. | Aug 2015 | A1 |
| 20150238091 | Iyer et al. | Aug 2015 | A1 |
| 20150238106 | Lappalainen et al. | Aug 2015 | A1 |
| 20150238147 | Figgatt et al. | Aug 2015 | A1 |
| 20150238148 | Georgescu et al. | Aug 2015 | A1 |
| 20150244946 | Agaian et al. | Aug 2015 | A1 |
| 20150245782 | Morland et al. | Sep 2015 | A1 |
| 20150246195 | Baker, Jr. et al. | Sep 2015 | A1 |
| 20150251012 | Olson | Sep 2015 | A1 |
| 20150257668 | Braojos Lopez et al. | Sep 2015 | A1 |
| 20150257671 | Laughner et al. | Sep 2015 | A1 |
| 20150257700 | Fu | Sep 2015 | A1 |
| 20150257712 | Sarrafzadeh et al. | Sep 2015 | A1 |
| 20150257715 | Quan et al. | Sep 2015 | A1 |
| 20150261926 | Kasabov | Sep 2015 | A1 |
| 20150265161 | Hernandez et al. | Sep 2015 | A1 |
| 20150265164 | Gopalakrishnan et al. | Sep 2015 | A1 |
| 20150265174 | Shakibi Gilani | Sep 2015 | A1 |
| 20150265175 | Shakibi Gilani | Sep 2015 | A1 |
| 20150265348 | Avitall et al. | Sep 2015 | A1 |
| 20150269825 | Tran | Sep 2015 | A1 |
| 20150272457 | Etemad et al. | Oct 2015 | A1 |
| 20150272464 | Armoundas | Oct 2015 | A1 |
| 20150283027 | Lampe et al. | Oct 2015 | A1 |
| 20150286779 | Bala et al. | Oct 2015 | A1 |
| 20150288573 | Baughman et al. | Oct 2015 | A1 |
| 20150289210 | Zhao et al. | Oct 2015 | A1 |
| 20150289777 | Linker | Oct 2015 | A1 |
| 20150289807 | Narayan et al. | Oct 2015 | A1 |
| 20150290453 | Tyler et al. | Oct 2015 | A1 |
| 20150297104 | Chen et al. | Oct 2015 | A1 |
| 20150297111 | Kassab | Oct 2015 | A1 |
| 20150297112 | Kassab et al. | Oct 2015 | A1 |
| 20150297907 | Zhang | Oct 2015 | A1 |
| 20150305632 | Najarian et al. | Oct 2015 | A1 |
| 20150320316 | Dhawan | Nov 2015 | A1 |
| 20150342478 | Galen et al. | Dec 2015 | A1 |
| 20150342488 | Laughner et al. | Dec 2015 | A1 |
| 20150342490 | Sachin et al. | Dec 2015 | A1 |
| 20150351699 | Addison et al. | Dec 2015 | A1 |
| 20150352367 | Quan et al. | Dec 2015 | A1 |
| 20150352369 | Quan et al. | Dec 2015 | A1 |
| 20150353880 | Clark et al. | Dec 2015 | A1 |
| 20150355459 | Shen et al. | Dec 2015 | A1 |
| 20150356350 | Awad | Dec 2015 | A1 |
| 20150359441 | Giovangrandi et al. | Dec 2015 | A1 |
| 20150359452 | Giovangrandi et al. | Dec 2015 | A1 |
| 20150359467 | Tran | Dec 2015 | A1 |
| 20150359486 | Kovacs et al. | Dec 2015 | A1 |
| 20150359492 | Giovangrandi et al. | Dec 2015 | A1 |
| 20150362360 | Kovacs et al. | Dec 2015 | A1 |
| 20150363108 | Kumar | Dec 2015 | A1 |
| 20150363193 | Chmiel et al. | Dec 2015 | A1 |
| 20150363194 | Chmiel et al. | Dec 2015 | A1 |
| 20150366511 | Addison et al. | Dec 2015 | A1 |
| 20150366518 | Sampson | Dec 2015 | A1 |
| 20150366532 | Voigt et al. | Dec 2015 | A1 |
| 20150374300 | Najarian et al. | Dec 2015 | A1 |
| 20150374983 | Simon et al. | Dec 2015 | A1 |
| 20160000346 | Lo et al. | Jan 2016 | A1 |
| 20160000350 | Zhang | Jan 2016 | A1 |
| 20160001089 | Allavatam et al. | Jan 2016 | A1 |
| 20160007907 | Lancaster et al. | Jan 2016 | A1 |
| 20160007932 | Laughner et al. | Jan 2016 | A1 |
| 20160008613 | Snyder | Jan 2016 | A1 |
| 20160022156 | Kovacs et al. | Jan 2016 | A1 |
| 20160022164 | Brockway et al. | Jan 2016 | A1 |
| 20160022166 | Stadler et al. | Jan 2016 | A1 |
| 20160022999 | Zhang et al. | Jan 2016 | A1 |
| 20160023013 | Greenhut et al. | Jan 2016 | A1 |
| 20160024156 | Barouch et al. | Jan 2016 | A1 |
| 20160033319 | Kovacs | Feb 2016 | A1 |
| 20160033622 | Martone et al. | Feb 2016 | A1 |
| 20160034634 | Hong et al. | Feb 2016 | A9 |
| 20160038037 | Kovacs | Feb 2016 | A1 |
| 20160038038 | Kovacs | Feb 2016 | A1 |
| 20160038091 | Krishnaswamy et al. | Feb 2016 | A1 |
| 20160041174 | Chinnaiyan et al. | Feb 2016 | A1 |
| 20160045123 | Bar-Tal et al. | Feb 2016 | A1 |
| 20160051203 | Furness et al. | Feb 2016 | A1 |
| 20160051205 | Al-Ali et al. | Feb 2016 | A1 |
| 20160051233 | Mo et al. | Feb 2016 | A1 |
| 20160051822 | Guez | Feb 2016 | A1 |
| 20160058301 | Shusterman | Mar 2016 | A1 |
| 20160058308 | Robinson | Mar 2016 | A1 |
| 20160066788 | Tran et al. | Mar 2016 | A1 |
| 20160066799 | Berkow | Mar 2016 | A1 |
| 20160066860 | Sternickel et al. | Mar 2016 | A1 |
| 20160066881 | Li et al. | Mar 2016 | A1 |
| 20160067433 | Martin et al. | Mar 2016 | A1 |
| 20160073313 | Hejazi et al. | Mar 2016 | A1 |
| 20160073959 | Eagle et al. | Mar 2016 | A1 |
| 20160073965 | Addison et al. | Mar 2016 | A1 |
| 20160074667 | Sullivan et al. | Mar 2016 | A1 |
| 20160081561 | Huang et al. | Mar 2016 | A1 |
| 20160081573 | Niebauer et al. | Mar 2016 | A1 |
| 20160087603 | Ricci et al. | Mar 2016 | A1 |
| 20160095531 | Narayan et al. | Apr 2016 | A1 |
| 20160098835 | Zhao et al. | Apr 2016 | A1 |
| 20160100787 | Leung et al. | Apr 2016 | A1 |
| 20160100803 | Korzinov et al. | Apr 2016 | A1 |
| 20160100817 | Hussain | Apr 2016 | A1 |
| 20160106378 | Kyal et al. | Apr 2016 | A1 |
| 20160113618 | Su et al. | Apr 2016 | A1 |
| 20160114162 | Sheldon et al. | Apr 2016 | A1 |
| 20160114168 | Demmer et al. | Apr 2016 | A1 |
| 20160114169 | Sheldon et al. | Apr 2016 | A1 |
| 20160120418 | Oksala et al. | May 2016 | A1 |
| 20160120431 | Habte et al. | May 2016 | A1 |
| 20160120433 | Hughes et al. | May 2016 | A1 |
| 20160120434 | Park et al. | May 2016 | A1 |
| 20160120464 | Lau et al. | May 2016 | A1 |
| 20160128630 | Warren et al. | May 2016 | A1 |
| 20160128641 | Fonseca et al. | May 2016 | A1 |
| 20160135705 | Liu et al. | May 2016 | A1 |
| 20160135754 | Marshall et al. | May 2016 | A1 |
| 20160136431 | Grill et al. | May 2016 | A1 |
| 20160140834 | Tran | May 2016 | A1 |
| 20160143543 | Zhang | May 2016 | A1 |
| 20160143594 | Moorman et al. | May 2016 | A1 |
| 20160148531 | Bleich et al. | May 2016 | A1 |
| 20160151013 | Atallah et al. | Jun 2016 | A1 |
| 20160152438 | Siikonen et al. | Jun 2016 | A1 |
| Number | Date | Country | |
|---|---|---|---|
| 20160361041 A1 | Dec 2016 | US |
| Number | Date | Country | |
|---|---|---|---|
| 62175686 | Jun 2015 | US |
