Patent Application
20240350098
The disclosed concept relates to medical alert systems, and, in particular, to a system and method for providing an alert of a fall risk. More particularly, in certain embodiments, the disclosed concept relates to a system and method for detecting abnormal blood pressure drops, and/or increase or decrease in pulse rate, in real-time (as described herein) using individualized criteria and providing actionable cues to mitigate a potential injurious fall.
According to the World Health Organization, falls are the second leading cause of accidental or unintentional injury deaths worldwide. It is estimated that 205 million falls occurred in 2015, with 37.3 million falls resulting in the need for medical attention and 646,000 falls resulting in death. Over 80% of the deaths are from low- and middle-income countries.
There are multiple reasons and causes for these falls. The disclosed concept addresses falls caused by abnormal blood supply to brain resulting in symptoms such as dizziness or even fainting. To illustrate the significance of this problem, fall statistics only due to orthostatic hypotension or OH (i.e. dizziness upon standing) is discussed. It is estimated that ⅓ of all adults 65 years of age and older experience OH, which places them at a risk for falling. Approximately 5% of people between the ages of 45 and 65 living in the community experience falls due to OH. In professional care settings (e.g., in hospitals, skilled nursing centers, assisted living environments, or other care facilities), the fall rate due to OH is as high as 64%.
To minimize injuries after a fall, attempts to provide prompt medical care is facilitated by the use of fall prevention devices such as bed, chair, and floor mat alarms that detect fall impacts. High fall risk individuals are frequently monitored by facility staff or a sitter who remains in the same room to supervise the individual. There are even solutions where a person at an off-site location monitors older adults via a PC screen and manually places calls to facilities notifying them of a high fall-risk situation (i.e. someone has made a motion to leave their bed). With the number of older adults expected to more than double by 2050 and to more than triple by 2100, rising from 962 million globally in 2017 to 2.1 billion in 2050 and 3.1 billion in 2100, these solutions are not adequate and not scalable.
Abnormal blood pressure drops and/or heart rate results in decreased blood flow to the brain resulting in individuals to experience fall risk symptoms such as light-headedness, visual blurring, dizziness, generalized weakness, fatigue, cognitive slowing, leg buckling, coat-hanger ache, and gradual or sudden loss of consciousness. When an individual assumes an upright position from a lying or sitting position, approximately 500 to 1,000 ml of blood pools to the legs and internal abdominal organs. Similar pooling occurs when an individual stands for a long period of time. In healthy individuals, the physiological response consists of an increase in heart rate and blood pressure so that the blood pooled in the lower extremities can be redistributed to vital organs including the brain, thereby avoiding fall risk symptoms. However, due to diseases and natural effects of aging, the physiological response of older adults can be delayed, inadequate, or non-existent, placing them at a high risk of an injurious fall.
Abnormal blood pressures and heart rate upon assuming an upright position or while in an upright position is defined by a progressive and sustained fall in systolic BP from baseline value ≥20 mmHg and/or diastolic BP≥10 mmHg, or an increase in heart rate of ≥30 beats per minute within three minutes of standing. This is illustrated in
Clinically, the inability for the body to compensate for orthostatic stress by regulating blood pressure and/or heart rate is a sign of poor baroreflex sensitivity and that the autonomic nervous system (ANS) of the individual is not working well. To elaborate, the ANS is located in the brain and it auto-regulates organs and nerves to keep an individual's body in continuous healthy balance, also known as homeostasis. ANS consists of two branches called the parasympathetic nervous system (PNS) and sympathetic nervous system (SNS). In healthy individuals at rest, both the SNS and PNS are active, with SNS being the dominant system. ANS dysfunction can be caused by or is a result of factors such as, but not limited to, natural age-related physiological decline, hypertension medication called vasodilators, decrease in blood volume due to dehydration, Parkinson's disease, and damaging effects on the autonomic system due to high levels of glucose in people with diabetes. Furthermore, individuals with impaired cardiac function due to structural heart disease, impaired cardiovascular adrenergic function (neurogenic OH), pure autonomic failure (PAF), multiple system atrophy (MSA), alcohol consumption, and exposure to heat have ANS dysfunction placing them at higher risk of experiencing fall risk symptoms.
Clinically, various diseases related to ANS dysfunction are diagnosed through the use of a tilt table and a procedure called the head-up tilt (HUT) test or a tilt table test (TTT). In such test, a person in supine position is strapped to a table and kept at rest for ≥5 minutes. Blood pressure and heart rate are measured and are used as baseline values. The individual is then rotated 60-70 degrees and kept passive in this position for 20 minutes to 40 minutes maximum. During this passive position, indications of abnormal blood pressures and/or heart rates signify various diseases related to ANS dysfunctions such as but not limited to OH, syncope, and postural orthostatic hypotension. The reported sensitivity, specificity, and reproducibility of such testing ranges greatly (e.g., from as low as 55% to as high as 96%). This large range is attributed to the fact that symptoms are transient, and that often results in individuals being misdiagnosed. When a person is under professional care and is susceptible to ANS dysfunction, s/he is placed under greater observation and told to be careful. When individuals are misdiagnosed or undiagnosed, this can compromise how they are treated medically as well as how they are supervised when they are admitted to facilities such as hospitals and various forms of long-term care. There is no current commercially available system to predict when someone would experience fall risk symptoms associated with ANS dysfunction (i.e. abnormal blood pressure and/or heart rate), thus driving the motivation for the disclosed concept described below.
In one embodiment, a method of predicting the risk of experiencing symptoms related to abnormal blood pressure and/or heart rate is provided. The method includes obtaining subject heart rate variability (HRV) data representing a number of HRV parameters, wherein the subject HRV data is generated based on heartbeat data obtained from an individual wearing a heart parameter sensor. The method further includes providing the subject HRV data as an input to an artificial intelligence system, wherein the artificial intelligence system has been previously trained using training and test HRV data representing the number of HRV parameters obtained from a plurality of test subjects. Finally, the method includes analyzing temporal data changes in or indicated by the subject HRV data in the artificial intelligence system to determine whether the individual is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate placing them at risk of a fall.
In another embodiment, an apparatus for predicting the risk of experiencing symptoms related to abnormal blood pressure and/or heart rate is provided. The apparatus includes a computer system comprising a number of controllers implementing an artificial intelligence system that has been previously trained using training and test heart rate variability (HRV) data representing a number of HRV parameters obtained from a plurality of test subjects. The artificial intelligence system is structured and configured to obtain subject HRV data representing the number of HRV parameters, wherein the subject HRV data is generated based on heartbeat data obtained from an individual wearing a heart parameter sensor, provide the subject HRV data as an input to the artificial intelligence system, and analyze temporal data changes in or indicated by the HRV data in the artificial intelligence system to determine whether the individual is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate.
In yet another embodiment, a method of training an artificial intelligence system to predict the risk of experiencing symptoms related to abnormal blood pressure and/or heart rate is provided. The method includes obtaining heartbeat data and blood pressure data from a plurality of test subjects, generating training and test heart rate variability (HRV) data representing a number of HRV parameters from the heartbeat data and the blood pressure data, and training the artificial intelligence system to predict risk of experiencing symptoms related to abnormal blood pressure and/or heart rate using the training and test HRV data based on temporal data changes in or indicated by the training and test HRV data.
In still another embodiment, a system for predicting the risk of experiencing symptoms related to abnormal blood pressure and/or heart rate is provided. The system includes a wearable biometric sensor including a heart parameter sensor structured and configured to generate heartbeat data from an individual wearing the wearable biometric sensor, and a computer system comprising a number of controllers implementing a predictive artificial intelligence system comprising an artificial intelligence system that has been previously trained using training and test heart rate variability (HRV) data representing a number of HRV parameters obtained from a plurality of test subjects. The predictive artificial intelligence system is structured and configured to obtain subject HRV data representing the number of HRV parameters, wherein the subject HRV data is generated based the heartbeat data, provide the subject HRV data as an input to the artificial intelligence system, and analyze temporal data changes in or indicated by the HRV data in the artificial intelligence system to determine whether the individual is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate.
In another embodiment, a method of outputting a warning signal when a subject is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate is provided. The method receiving heart rate variability data of a subject wherein the HRV data represents at least one HRV parameter, wherein the person's HRV data is generated based on heartbeat data obtained from a heart parameter sensor worn by the person, and receiving motion data of the person from a motion sensor worn by the person, the motion data being sufficient to distinguish between resting data and vertical rising data, the resting data representing when the subject is resting in the lying or sitting position and the vertical rising data representing when the subject is moving from a lying position to a sitting or standing position or from a sitting position to a standing position, the resting data and the vertical rising data being synced to the HRV data to identify a portion of the HRV data representative of when the person is resting and a portion of the HRV data representative of when the person is vertically rising. The method further includes determining from temporal data changes in or indicated by the subject HRV data associated only with the portion of the HRV data representative of when the person is resting and optionally, the portion of the HRV data representative of when the person is vertically rising whether the person is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate, and outputting the warning signal when the determining step determines the person is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate.
In still another embodiment, an apparatus worn by a subject in a resting position to provide a warning signal when the subject is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate is provided. The apparatus includes a heart rate sensor capable of measuring an interbeat interval at least 100 hertz, the heart rate sensor being attachable to a body of the subject, and a computer in communication with the heart rate sensor. The computer is operative to analyze the data of the heart rate sensor to output the warning signal when the subject is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate based solely only on the data from the heart rate sensor relating to when the subject is in the resting position.
In yet another embodiment, an apparatus worn by a subject to provide a warning signal when the subject is at risk of falling is provided. The apparatus includes a heart rate sensor capable of measuring an interbeat interval of at least 100 hertz, the heart rate sensor being attachable to a body of the subject, a motion sensor capable of sensing motion data for distinguishing when the subject is resting and when the subject is vertically rising from a lying position to a sitting or standing position or from the sitting position to the standing position, and a computer in communication with the heart rate sensor and the motion sensor and operative to analyze and output the warning signal as to when the subject is at risk of falling based at a minimum, on the data from the heart rate sensor and the motion sensor relating to when the subject is resting and optionally when the subject is rising from the lying position to the sitting or standing position or from the sitting position to the standing position.
In a further embodiment, a method of outputting a warning signal when a subject is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate is provided. The method includes the steps of receiving heart rate variability (HRV) data of a subject wherein the HRV data represents at least one HRV parameter, wherein the subject's HRV data is generated based on heartbeat data obtained from a heart parameter sensor worn by the subject, determining from temporal data changes in or indicated by the subject HRV data associated only with a portion of the HRV data representative of when the subject is resting and optionally, with a portion of the HRV data representative of when the subject is vertically rising whether the subject is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate, and outputting the warning signal when the determining step determines the subject is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate.
In a still further embodiment, a pedal pulse sensor device structured to wrap around a foot of a wearer is provided. The pedal pulse sensor device includes a first laterally extending portion, a second portion that extends transversely from the first portion, wherein at least one of the first portion and the second portion includes a cavity structured to be located along the dorsalis pedis or the posterior tibial arteries of the wearer responsive to the pedal pulse sensor device being wrapped around the foot of the wearer, and a biometric sensor unit held within the cavity, wherein the biometric sensor unit includes a heart parameter sensor.
As used herein, the singular form of “a”, “an”, and “the” include plural references unless the context dictates otherwise.
As used herein, the statement that two or more parts or components are “coupled” shall mean that the parts are joined or operate together either directly or indirectly, i.e., through one or more intermediate parts or components, so long as a link occurs.
As used herein, the term “number” shall mean one or an integer greater than one (i.e., a plurality).
As used herein, the term “random forest” shall mean machine learning decision trees for classification and/or regression. The collection of the decision trees makes up the forest with larger number of decision trees yielding higher accuracy results unless the context dictates otherwise.
As used herein, the term “deep learning neural network” shall mean an artificial neural network with multiple hidden layers between the input and output layers that determines the correct mathematical manipulation (linear or non-linear) to turn the input into the output by moving through the layers and calculating the probability of each output unless the context dictates otherwise.
As used herein, the term “hidden layer” shall mean a neural network layer of one or more neurons whose output is connected to the inputs of other neurons and that, as a result, is not visible as a network output unless the context dictates otherwise.
As used herein, the term “recurrent neural network” shall mean a class of artificial neural network where connections between nodes form a directed graph along a temporal sequence and that therefore allows the network to exhibit temporal dynamic behavior unless the context dictates otherwise.
As used herein, the term “recursive deep learning neural network” shall mean a deep learning neural network that is also a recurrent neural network unless the context dictates otherwise.
As used herein, the term “heart rate variability (HRV)” shall mean the variation in the time interval between consecutive heartbeats. In an ECG wave signal as shown in
As used herein, the term “HRV parameter” shall refer to statistical values derived from HRV data (e.g., RR intervals (although QQ and TT could also be used)), such as, but not limited to, averages or standard deviations. These HRV parameters shall include, without limitation, measures of HRV obtained using time-domain methods, frequency-domain methods, or nonlinear methods (e.g., a Poincaré plot).
As used herein, the term “low frequency (LF) range” shall mean from 0.04 to 0.15 Hz.
As used herein, the term “high frequency (HF) range” shall mean from 0.15 to 0.4 Hz.
As used herein, the term “controller” shall mean a programmable analog and/or digital device (including an associated memory part or portion) that can store, retrieve, execute and process data (e.g., software routines and/or information used by such routines), including, without limitation, a field programmable gate array (FPGA), a complex programmable logic device (CPLD), a programmable system on a chip (PSOC), an application specific integrated circuit (ASIC), a microprocessor, a microcontroller, a programmable logic controller, or any other suitable processing device or apparatus. The memory portion can be any one or more of a variety of types of internal and/or external storage media such as, without limitation, RAM, ROM, EPROM(s), EEPROM(s), FLASH, and the like that provide a storage register, i.e., a non-transitory machine readable medium, for data and program code storage such as in the fashion of an internal storage area of a computer, and can be volatile memory or nonvolatile memory.
As used herein, the terms “component” and “system” are intended to refer to a computer related entity, either hardware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to being, a process running on a processor, a processor, an object, an executable, a thread of execution, a program, and/or a computer. By way of illustration, both an application running on a server and the server can be a component. One or more components can reside within a process and/or thread of execution, and a component can be localized on one computer and/or distributed between two or more computers.
Directional phrases used herein, such as, for example and without limitation, top, bottom, left, right, upper, lower, front, back, and derivatives thereof, relate to the normal sense of the words.
The disclosed concept will now be described, for purposes of explanation, in connection with numerous specific details in order to provide a thorough understanding of the subject innovation. It will be evident, however, that the disclosed concept can be practiced without these specific details without departing from the spirit and scope of this innovation.
As described in greater detail herein in connection with particular exemplary embodiments, the disclosed concept deviates from current clinical practice of comparing pre- and post-orthostasis blood pressure and/or heart rate data to determine one's propensity of experiencing symptoms related abnormal blood pressures and/or heart rate upon orthostasis, such as, without limitation, OH symptoms.
In the disclosed concept, real-time heartbeat data, such as ECG QRS wave-form data or PPG sensor data, obtained during periods of low physical activity as indicated by physical motion sensors is extracted and selected features (HRV parameters) are analyzed by an artificial intelligence system. In one particular embodiment, both HRV parameters and the ECG QRS wave-form or PPG sensor data may be analyzed in combination to increase accuracy. Real-time may refer to the time period immediately before the person rises as detected by an accelerometer or other device for about the past 1 second to 1 hour, and preferably the past 2 seconds to 5 minutes.
The disclosed concept is able to determine and predict if an individual's ANS system is in a compromised state and unable to cope with the stress of standing. This is accomplished by using a device to detect when the individual is vertically rising, a heart rate monitor and an artificial based system. The accelerometer is discussed further below and is optional. The heart rate monitor is also discussed further below. The artificial intelligence (e.g., machine learning) based system may utilize, for example, an artificial neural network or random forest system, for detecting and providing alerts of potential symptoms related to abnormal blood pressure(s) and/or heart rate based on the real time monitoring of certain predetermined HRV parameters obtained from data collected by an ECG sensor, a PPG sensor, or other similar sensor device. In particular, as described in more detail herein, an artificial intelligence based system is trained to examine temporal changes in certain HRV parameters (determined from heartbeat data obtained before the individual stands up—e.g., as identified by certain physical motion sensors) in order to predict from such parameters and the temporal change thereof the risk that an individual will experience symptoms related to abnormal blood pressure(s) and/or heart rate if he or she stands. Temporal may mean past 1 second to 10 minutes, and preferably past 2 seconds to 5 minutes. The heartbeat data may be obtained by the person wearing a heart rate monitor. As an aspect of the disclosed concept, and as described in more detail herein, HRV parameters from human subject ECG or PPG test data are labeled with classifications (such as, but not limited to, whether or not abnormal blood pressure(s) and/or heart rate were present) and are used to train the artificial intelligence system. Specifically, the artificial intelligence system is trained so as to establish a set of baseline fall risk criteria and provide real-time prediction of fall risk episodes based on certain temporal changes of HRV parameter inputs. In addition, in a further aspect of the disclosed concept, the artificial intelligence system may be further trained to establish customized risk criteria for a particular individual using additional test data obtained specifically from that individual during one or more subsequent training phases.
Thus, in short, parameters established from, for example, ECG QRS wave-form make up the training data and are used to train an artificial intelligence algorithm to recognize abnormal blood pressure(s) and/or heart rate risks (e.g., OH risks) and/or intentions of standing. As discussed, the training data has been labeled with known classifications (e.g. FALLRISK, NO_FALLRISK, indicated by heart rate, systolic and diastolic blood pressures) and the supervised artificial intelligence (e.g., machine learning) algorithm learns to make predictions on the FALLRISK or NO_FALLRISK from new input data. In the exemplary embodiment, time series data (e.g., any HRV parameter versus time) is provided in short (e.g., 1 second) windows, although smaller or larger time windows can be used. Multiple parameters are combined, creating a rich time series that can be used by the algorithm to determine the best predictive model. Classifications of FALLRISK or NO_FALLRISK can be made using various statistical models (regression, Naïve Bayes, or Bayesian Networks) or using structural models that are rule based (Decision Trees, Random Forest), distance based (k-Nearest Neighbor, learning vector quantization), and Neural Networks (Multi-Layer Perceptron). In the exemplary embodiment, the Random Forest model is used, which implements ensemble theory to create a collection of decision tree (i.e., a forest) classifiers using randomly selected subsets of the training data. The model selects the best class to yield the highest predictive accuracy.
The disclosed concept is able to predict in advance if the person will be at FALLRISK or NO_FALLRISK (e.g., between 5 to 30 seconds in advance of it happening). Thus, care staff can opt to receive notifications as soon as the person is at risk. Another option is to allow the algorithm to send notifications as soon as the person has intentions to stand. Frail older adults typically require 15 to 30 seconds to transition from lying to standing. It is therefore anticipated that the disclosed concept will be used by frail older adults under professional care, and thus the ability to predict 5 to 30 seconds in advance is significant and allows the care staff to administer aide to the older adult in a timely fashion.
Thus, according to an aspect of the disclosed concept, described in greater detail herein in connection with a number of particular exemplary embodiments, a biometric sensor unit, such as a wearable health tracker, may be used to determine a person's RR interval (i.e., detect heartbeat data by, for example, single or multiple lead ECG methods). The biometric sensor unit can, for example, and without limitation, be worn on the chest as a strap, be worn on the wrist as a health watch, be worn in the car, be integrated into a piece of clothing, or even be implanted in the body (essentially anywhere a pulse can be detected). The RR (or NN) intervals are then used to determine certain HRV parameter inputs, which inputs are then (without concurrently measured blood pressure data) fed to and analyzed by the trained artificial intelligence system to determine whether certain criterion for fall risks have been met. The inputs can be continuous or intermittent (such as every 1 seconds or 3 seconds or 10 seconds or 30 seconds). In one embodiment, transmission from the biometric sensor unit is initiated when there is physical motion as indicated by a physical motion sensor to conserve battery life of the biometric sensor unit. The artificial intelligence system can process entire streams of data, but to increase the artificial intelligence system's computation speed, analysis can be completed on selective time instances where a physical motion sensor indicates low activity (i.e. lying, sitting, or standing) when an individual is at the highest risk of experiencing symptoms related to abnormal blood pressure and/or heart rate. If a determination is made that certain criterion for fall risk have been met, a real-time alert (controlled by the continual assessment of the individual's fall risk as described herein), such as a cell phone vibration, a sound, an image, and or a video, is triggered and provided to the individual. This alert provides actionable cues that serve as reminders for the individual to flex their ankles back and forth to manually pump blood in their lower extremities or to pause and refrain postural change until there is no fall risk as determined by the trained artificial intelligence system. A different alert may also be provided to a family member or care provider, notifying them of the individual's fall risk should they standup, so that preemptive actions can be taken. The real-time alert will be terminated once it is safe for the individual to resume changes in position as determined by the trained algorithm. In some embodiments, described in greater detail herein, the real-time detection of a change in body position is used in combination with the determination that certain criterion for fall risk have been met to cause the alert to be triggered.
As stated elsewhere herein, HRV parameters may be time-domain based, frequency-domain based and/or non-linear based, and many such parameters may be employed in connection with the implementation of the disclosed concept. Thus, before describing particular embodiments of the disclosed concept in detail, a description of various suitable HRV parameters that may be employed in the disclosed concept will first be provided.
Typical time-domain HRV parameters are derived from the RR or NN intervals of collected heartbeat data (e.g., ECG data). A number of such time-domain HRV parameters are shown in Table 1 below (each of which is a direct and indirect measure of RR or NN distribution).
As described in more detail herein, exemplary embodiments of the disclosed concept use one or more of these time-domain HRV parameters. It will be appreciated, however, that this is meant to be exemplary only, and that other time-domain HRV parameters (including those listed in Table 1 or additional parameters not listed in Table 1) may also be used within the scope of the disclosed concept. In one particular exemplary embodiment, three particular time-domain HRV parameters are used. Those three time-domain HRV parameters are pNN50, which represents the percentage of successive RR (or NN) intervals that differ by more than 50 ms, RMSSD, which represents the root mean square of successive RR (or NN) interval differences, and TINN, which is the baseline width of the RR (or NN) interval histogram. These variables are directly related to the short-term variation of the RR (or NN) intervals and corresponding high frequency content of the RR (or NN) intervals, and thus are believed to be particularly suited for providing information relating to the risk for experiencing symptoms related to abnormal blood pressure and/or heart rate and the ability to properly respond thereto.
Any time domain curve, in this case the RR interval versus time curve, can be represented as a sum of multiple sinusoidal oscillations at different amplitudes and frequencies between 0.0033 Hz to 0.4 Hz. The process of analyzing a time domain curve using frequencies is called the Fast Fourier Transform (FFT). Results of an FFT provide information on the frequency content of the curve and associated power of those frequencies.
The frequency bands are typically ultra-low frequency (ULF) from 0 to 0.0033 Hz, very low frequency (VLF) from 0.0033 to 0.04 Hz, low frequency (LF) from 0.04 to 0.15 Hz, and high frequency (HF) from 0.15 to 0.4 Hz. These bands are further divided into smaller bins, yielding finer details about the presence and power of these frequencies in the original RR interval versus time curve. The time varying nature of the frequency content of an individual's RR intervals can best be visualized using a 3D frequency spectrogram as shown in
In the exemplary embodiment, non-linear based domain analysis consists of creating a scatter plot called a Poincaré plot. In a Poincaré plot, the x-axis represents the RRn (or NNn) value with the subsequent RRn+1 (or NNn+1) value being plotted on the y-axis. For a given time window, a cluster of data points can be fitted with a rotated ellipse as shown in the exemplary Poincaré plot of
A number of particular examples of HRV parameters in accordance with the disclosed concept will now be discussed in connection with
After Test 1 was completed (900 seconds in supine (only 600 seconds is shown) and 300 seconds in standing position), the same individual conducted Test 2 (immediately after Test 1 without interruption).
Referring to
During each of the orthostatic stand tests, the individual first lies supine for 5 minutes to bring his/her body to a state of rest before each test. Then, the individual lies supine for 10 minutes, gets up from the table, and stands for 5 minutes. In this particular exemplary embodiment, this is done three times without interruption. Note, however, that this is meant to be exemplary only, and that other protocols for collecting data may also the used. For example, the same test can be conducted while sitting for 10 minutes and then standing for 5 minutes should the individual be limited in his/her ability to conduct the maneuver independently.
Next, at step 10, a number of HRV parameters (e.g., time-domain based, frequency-domain based and/or non-linear based parameters) are determined from the collected heartbeat data. Also, the collected blood pressure data is analyzed to identify abnormal blood pressure and/or heart rate by detecting sustained blood pressure drops or heart rate increases as described herein. Then, the analyzed blood pressure data is used to tag the HRV parameter data. More specifically, in one particular exemplary embodiment, when there is a systolic blood pressure drop(S), diastolic blood pressure drop (D), or increase in heart rate (H) that meets the definition of an abnormal blood pressure and/or heart rate episode (such as, without limitation, an OH episode), HRV parameter data for 0 to 30 seconds prior to standing up (or could be in the 30 seconds to minutes range, depending upon training results) is classified with an identifier. For example, if systolic blood pressure dropped, diastolic blood pressure did not drop, and heart rate increased, then the HRV parameter data prior to standing up is classified as FALLRISK_SH. If there are no signs of abnormal blood pressure and/or heart rate, then data prior to standing is tagged as NO FALLRISK. Thus, following step 10, data will have been assembled for each of the conducted orthostatic stand tests that includes classified or labelled HRV parameters for the time 0 to 30 seconds prior to standing up (or could be in the 30 seconds to minutes range, depending upon training results).
In one particular exemplary embodiment, the only HRV parameters that are determined at step 10 are frequency-domain based and comprise PSDs in the HF range or the LF and HF ranges. In another, alternative particular exemplary embodiment, time-domain based, frequency-domain based and/or non-linear based HRV parameters are determined at step 10 and include one or any combination of pNN50, RMSSD, and TINN as time-domain based HRV parameters, PSDs in the HF range or the LF and HF ranges as frequency-domain based HRV parameters, and Poincaré plots and/or or one or more parts thereof (e.g., the SD1 and SD2 values as described herein) as non-linear based HRV parameters. Other suitable parameters include average duration and rate of change of Q to R and R to S of the ECG QRS waveform. The QRS waveform represents the electrical signal of the heart as the ventricles begin to contract (point Q), where the bulk of the ventricle muscles completes its contraction (point R) and marks the beginning of systole. That is, the start of the heart pumping blood into the arteries. There are smaller areas of the ventricle that contract and complete the full contraction cycle, and are identified by point S. Again, it will be understood that these particular embodiments are meant to be exemplary only, and that other implementations using different sets of parameters from the QRS waveform may also be used within the scope of the disclosed concept. For example, and without limitation, the following parameters in any combination may be used alone or in combination with the particular embodiments just described: PNS index (which is a function of mean RR, RMSSD, and SD1) and SNS index (which is a parameter based upon mean heart rate, Baevsky's stress index and SD2), average duration and rate of change of Q to R and/or R to S, QQ intervals, RR intervals, SS intervals, and heart rate.
Next, at step 15, the determined one or more parameters derived from the QRS waveform and the analyzed (e.g., labeled) blood pressure data are used to train and test the artificial intelligence system. Specifically, the artificial intelligence system, which is a machine learning system (e.g., artificial neural network) in the exemplary embodiment, is trained to be able to predict abnormal blood pressure and/or heart rate episodes (and, in particular, when an individual's ability to respond to orthostatic stress is compromised) based on the selected HRV parameter inputs, and in particular based on the monitoring of temporal changes in those HRV parameter inputs. As a result, the trained artificial intelligence system will establish and be able to detect a set of baseline criteria for predicting abnormal blood pressure and/or heart rate episodes based on certain QRS waveform parameter inputs. As noted elsewhere herein, in an alternative exemplary embodiment, once deployed, the artificial intelligence system may thereafter be further trained with data specific to a particular individual in question by collecting additional heartbeat and blood pressure data from that individual during one or more subsequent training phases, and using that collected data to further train the artificial intelligence system so as to be customized to that particular individual.
Thus, following step 15, an artificial intelligence system is provided that is able to predict when an individual's ability to respond to orthostatic stress is compromised. In other words, the trained artificial intelligence system has the ability to recognize and predict when a person's autonomic nervous is incapable of coping with orthostatic stress leading the person to experience symptoms related to abnormal blood pressure and/or heart rate. As discussed elsewhere herein, these symptoms include, but are not limited to, light-headedness, visual blurring, dizziness, generalized weakness, fatigue, cognitive slowing, leg buckling, coat-hanger ache, and gradual or sudden loss of consciousness (i.e. syncope). These symptoms place the individual at a high risk of an injurious fall. As just described, this is accomplished by analyzing temporal changes in the parameters derived from heartbeat (e.g., ECG) data, specifically the RR or NN interval data, in the time, frequency, and/or nonlinear domains.
Biometric sensor unit 25 is structured and configured to be worn by an individual to be monitored. For example, biometric sensor unit 25 may be worn by an individual at a hospital, nursing home, or any other location where the individual might be at a risk of falling and therefore needs to be monitored.
Biometric sensor unit 25 further includes a controller 70 coupled to receive the outputs of heart parameter sensor 55. Finally, wearable biometric sensor unit 25 includes a short-range wireless communications module 75 that is structured and configured to enable wearable biometric sensor unit 25 to communicate with receiver unit 30 over a short-range wireless network. Short-range wireless communications module 75 may be, for example and without limitation, a WiFi module, a Bluetooth® module, a ZigBee module, an IEEE802.15.4 module, or any other suitable short-range wireless communications module that provides compatible communications capabilities.
Referring again to
Network 35 may be, for example, the Internet, one or more private communications networks, or any combination thereof. As employed herein, the term “communications network” shall expressly include, but not be limited by, any local area network (LAN), wide area network (WAN), intranet, extranet, global communication network, the Internet, and/or wireless communication network. Preferably, the wired and/or wireless connections to network 35 are secure (e.g., in the form of an encrypted virtual private network).
Central computer system 40 comprises any suitable processing or computing system having a computing device and one or more memory components for data storage (e.g., a controller), such as, without limitation, one or more PCs or server computers. As seen in
In operation, heart parameter sensor 55 of biometric sensor unit 25 continuously collects heartbeat data (e.g., ECG data) from the individual wearing biometric sensor unit 25 and transmits that data wirelessly to receiver unit 30 using short-range wireless communications module 75. In the exemplary embodiment, the collected heartbeat data is transmitted to receiver unit 30 in packets of time which can range from seconds to minutes. Receiver unit 30 in turn transmits the received heartbeat data to central computer system 40 through network 35 for analysis by predictive AI system 45 of central computer system 40. Note that in the present implementation, a blood pressure monitor is no longer needed once the algorithm of predictive AI system 45 has been trained. However, it could be used as a redundant system.
Predictive AI system 45 receives the heartbeat data from receiver unit 30 as just described and uses the trained artificial intelligence system of predictive AI system 45 to predict the likelihood of the onset of experiencing symptoms related to abnormal blood pressure and/or heart rate by checking the heartbeat data against the baseline criteria. More specifically, predictive AI system 45 determines from the received heartbeat data those particular HRV parameters that have been used to train the artificial intelligence system of predictive AI system 45 as described elsewhere herein. Predictive AI system 45, in particular the trained artificial intelligence system thereof, then analyzes the determined HRV parameters to determine the level of fall risk based on the baseline criteria. In the exemplary embodiment, the risk is calculated in terms of a percentage. If predictive AI system 45 determines that the criteria are met (e.g., if greater than a predetermined percentage risk is determined), central computer system 40 will generate and transmit a signal to alert a care provider (e.g. professional care staff, family, and/or loved ones) and the individual themselves in advance of the impending onset of experiencing symptoms related to abnormal blood pressure and/or heart rate so preemptive actions can be taken to mitigate potential injuries. In particular, in one embodiment, central computer system 40 generates and transmits a signal to care provider computer system 50 (facilitated through web applications) in order to provide advance notice of the fall risk associated with the impending onset of experiencing symptoms related to abnormal blood pressure and/or heart rate for the individual wearing biometric sensor unit 25. In the exemplary embodiment, display 85 of care provider computer system 50 will display information identifying the current fall risks at the location in question as shown in
In the exemplary embodiment, based upon the risk level determined by predictive AI system 45, bedside biofeedback monitor 95 will output certain visual, tactile and/or audible information. For example, as shown in
In still a further alternative embodiment, to increase accuracy, the artificial intelligence system of predictive AI system 45 is trained with patient specific heartbeat data (e.g., ECG data) to establish individualized criteria for experiencing symptoms related to abnormal blood pressure and/or heart rate. More specifically, the artificial intelligence system of predictive AI system 45 in
More specifically, in this exemplary embodiment, physical motion parameters derived from data collected from a pool of test subjects (e.g., by way of sensor(s) similar to physical motion sensor(s) 60) are used to train predictive AI system 45 to be able to determine/predict when an individual has an intention of standing. Once so trained, predictive AI system 45 may thereafter be used to determine/predict when a wearer of biometric sensor unit 25″ has an intentions of standing based on data collected by physical motion sensors 60.
The particular physical motion parameters that may be used include, for example and without limitation, the rate of change, absolute magnitudes, or difference magnitudes in each axis direction of certain sensed data, and/or statistical descriptors of such data, such as mean, standard deviation, and variance of these parameters. In addition, in the exemplary embodiment, the training data includes data that is labeled with the start of specific body movements. Such labels may include, for example, lying on the left side, lying on the right side, lying on the back, or lying on stomach. In addition, the data is further labeled with the instant that physical position changes are made. Different physical maneuvers may also be identified. For example, such maneuver may include an individual maneuvering from lying to sitting by making a sit-up exercise motion, or maneuvering from lying to sitting by rolling onto their side (left or right), and then pushing off the bed to an upright sitting position.
Thus, in connection with the various embodiment described herein (including those employing an altimeter as described below), the motion data generated by physical motion sensors 60 will be sufficient to distinguish between resting data and vertical rising data, wherein the resting data represents when the subject is resting in the lying or sitting position and the vertical rising data represents when the subject is moving from a lying position to a sitting or standing position or from a sitting position to a standing position. In addition, the resting data and the vertical rising data will be synced to the HRV data to identify the portion of the HRV data that is representative of when the person is resting and the portion of the HRV data that is representative of when the person is vertically rising.
In addition, a low- and high-pass filter may be used to separate the sensed time varying signal into AC and DC components. The AC component is generally attributed to accelerations of the body and can be used to detect and predict intentions of standing. The DC component is the gravitational contribution and can also be used to classify body postures (lying, sitting, standing, etc.).
In the exemplary embodiment, predictive AI system 45 is first trained with data in order to establish a baseline set of criteria as described elsewhere herein. Thereafter, predictive AI system 45 may be further trained to establish individual-specific physical motion intentions using data from the individual using physical motion sensor(s) 60.
In one exemplary embodiment, physical motion sensor(s) 60 consist of a 3-axis accelerometer. In another exemplary embodiment, physical motion sensor(s) 60 consists of a 3-axis accelerometer and a 3-axis gyroscope. In still another exemplary embodiment, physical motion sensor(s) 60 consists of a 3-axis accelerometer, a 3-axis gyroscope, and a 3-axis magnetometer. In yet other embodiments, physical motion sensor(s) 60 comprise a laser beam, radar, and/or proximity sensor, in any combination and/or in combination with the sensor(s) described above or below.
In a minimized configuration, intentions of standing may be predicted using data from only 1-axis of an accelerometer and 1-axis of a gyroscope without the use a magnetometer. Use of this minimized configuration allows determination of the angle of tilt of a person's body with respect to the vertical direction of gravity. Although drift in gyroscope signal are typical, it will likely be minimal in terms of predicting intentions of standing which happens within seconds rather than minutes or hours.
In still another exemplary configuration, physical motion sensor(s) 60 comprise a 9-axis inertial motion unit (IMU). Data collected by the IMU is from a 3-axis accelerometer to detect acceleration (e.g., g-forces) in the x-, y-, and z-directions, a 3-axis gyroscope to detect rotation of a person's body (e.g. roll, pitch, and yaw about the three axes), and a 3-axis magnetometer that measures the Earth's magnetic field and acts like a compass to correct signal drift of the gyroscope signal minimizing error in absolute direction of the individual can also be used. Typical acceleration (e.g., g-force) of a person standing is within +/−2 g range. In the exemplary embodiment, the gyroscope is at a minimum able to detect body rotation of 250 degrees per second, and the magnetometer is be able to detect changes in the magnetic field of 1300 μT. In some implementations, the data from all of the sensors in the IMU needs to be filtered for noise, errors, and drift. This step, however, is unnecessary in some IMUS (such as the commercially available BOSCH BNO055 IMU), as those units fuse all of the sensor outputs together and output a final fusion result vector suitable for training an artificial intelligence system such as predictive AI system 45.
During operation, biometric sensor unit 25″ is, in one embodiment, located on the center of the chest. The chest is preferable as it is the part of the body that has limited motion during sedentary activities (lying or sitting). However, biometric sensor unit 25″ could, during operation, be placed anywhere on the individual, such as the car, shoulders, wrist, leg, foot, garments such as socks, or on the medical wrist bands.
Any of the biofeedback devices shown in
As discussed elsewhere herein, a PPG sensor may be employed to collect heartbeat data for use as described herein. PPG sensors are usually placed in devices that are worn on the wrist, chest, and cars. However, in still a further particular alternative embodiment of the disclosed concept, an alternative location is used for the PPG sensor along the dorsalis pedis and/or posterior tibial arteries available in the lower extremities of the leg as shown in
Determination of RR (or NN) intervals using PPG sensors is challenging as it is prone to motion artifacts that deteriorate the training data. Thus, a location, such as the lower leg, where there is minimal movement during sedentary activities (lying, sitting) is desirable. The locations identified by the present inventor to acquire PPG data are along the dorsalis pedis and posterior tibial arteries and they are marked with an “X” in
More specifically, as seen in
In use, pedal pulse sensor device 200 will be wrapped around the foot. Once wrapped around the foot, Location A will meet Location C and Location D will meet Location B (with the hook and loop fasteners as described providing the means for securing to the foot as shown). In the exemplary embodiment, pedal pulse sensor device 200 will be offered in different sizes (e.g., small, medium, large, and extra-large).
In one particular embodiment, pedal pulse sensor device 200 is structured and configured with the capability to detect when an individual's foot has transitioned from the bed and has made a descent towards the floor. To do so, one of the physical motion sensors in biometric sensor unit 25″ held within pedal pulse sensor device 200 is a micro altimeter pressure sensor, such as Servoflo Corporation MS5611-01BA, capable of measuring change in altitude as small as 10 cm. Alternatively, a similar device may be integrated into a sock or worn on the sock either as a clip, snap, or strap, or placed into a cavity sewn into the sock. Accelerometers, gyroscopes, and magnetometers could also be integrated into the system as described herein, with data from these sensors used training the artificial intelligence system to predict when a person has intentions of leaving their bed.
In yet another particular embodiment, a system and method are provided, by way of modification to any of systems 20, 20′ or 20″, for predicting the risk of experiencing symptoms related to abnormal blood pressure and/or heart rate. In this embodiment, subject heart rate variability (HRV) data representing a number of HRV parameters is received, wherein the subject HRV data is generated based on heartbeat data obtained from an individual wearing a heart parameter sensor (such as biometric sensor units 25 and/or biometric sensor units 25′), but only while the individual is in a lying or sitting position prior to standing up. Temporal data changes in or indicated by the received subject HRV data are then analyzed (e.g., by predictive AI system 45) to determine therefrom whether the individual is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate. If it is determined that the individual is at risk of experiencing symptoms related to abnormal blood pressure and/or heart rate, an output signal is generated that is indicative of the risk level. In this embodiment, the determination as to whether the heartbeat data is only from a period where the individual is in a lying or sitting position prior to standing up is based on motion data collected by a number of physical motion sensors (e.g., sensor(s) 60) worn by the individual.
In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word “comprising” or “including” does not exclude the presence of elements or steps other than those listed in a claim. In a device claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The word “a” or “an” preceding an element does not exclude the presence of a plurality of such elements. In any device claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The mere fact that certain elements are recited in mutually different dependent claims does not indicate that these elements cannot be used in combination.
Although the invention has been described in detail for the purpose of illustration based on what is currently considered to be the most practical and preferred embodiments, it is to be understood that such detail is solely for that purpose and that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover modifications and equivalent arrangements that are within the spirit and scope of the appended claims. For example, it is to be understood that the present invention contemplates that, to the extent possible, one or more features of any embodiment can be combined with one or more features of any other embodiment.
This is a divisional application which claims priority from U.S. patent application Ser. No. 17/053,425, filed on Nov. 6, 2020, entitled “SYSTEM FOR MONITORING AND PROVIDING ALERTS OF A FALL RISK BY PREDICTING RISK OF EXPERIENCING SYMPTOMS RELATED TO ABNORMAL BLOOD PRESSURE(S) AND/OR HEART RATE”, which is a U.S. National Stage Application under 35 U.S.C. § 371 of International Application No. PCT/US2019/031042, filed on May 7, 2019, entitled “SYSTEM FOR MONITORING AND PROVIDING ALERTS OF A FALL RISK BY PREDICTING RISK OF EXPERIENCING SYMPTOMS RELATED TO ABNORMAL BLOOD PRESSURE(S) AND/OR HEART RATE”, which claims priority under 35 U.S.C. § 119 (c) from U.S. Provisional Patent Application No. 62/668,428, filed on May 8, 2018, entitled “ORTHOSTATIC HYPOTENSION ALERT SYSTEM”, the contents of which are incorporated herein by reference.
| Number | Date | Country | |
|---|---|---|---|
| 62668428 | May 2018 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 17053425 | Nov 2020 | US |
| Child | 18762823 | US |
