Methods and apparatus for monitoring cardiovascular regulation using heart rate power spectral analysis

BACKGROUND OF THE INVENTION
The present invention relates in general to methods and apparatus for monitoring cardiovascular regulation and in particular to methods and apparatus for heart rate spectral analysis.
Changes in cardiovascular regulation associated with congestive heart failure include attenuation of activity in the parasympathetic division of the autonomic nervous system, enhancement of activity in the sympathetic division of the autonomic nervous system, cardiac catecholamine depletion, down regulation of the beta-receptor system, increased renin-angiotensin system activity, and alteration of baroreceptor function. All of these regulatory changes require either specific clinical manipulations, such as a stress test, a Valsalva maneuver, or the like, and/or invasive maneuvers, such as cardiac biopsy, plasma catecholamine measurement, or the like, in order to determine the extent of regulatory dysfunction and its impact upon the clinical state of the patient and upon prognoses for the patient. These procedures are time consuming, and generally do not permit the formation of a clinical judgment and subsequent action within the timeframe of the course of treatment for critically ill patients in an Intensive Care Unit.
Fluctuations from heartbeat to heartbeat in measured properties of the circulatory system reflect both the presence of a variety of naturally occurring physiological disturbances of the circulatory system homeostasis, and the dynamic response of cardiovascular control systems to these disturbances. For example, the cyclic variation in intrathoracic pressure which accompanies breathing mechanically affects the return of venous blood to the heart and also affects blood pressure in pulmonary vessels and in the aorta. The variation in intrathoracic pressure is also coupled to a cyclic variation in heart rate through a neural mechanism mediated by the central nervous system. Furthermore, the resulting cyclic variation in arterial blood pressure impinges on heart rate through a reflex, known as the baroreceptor reflex, which is mediated by the autonomic nervous system. Disturbances in cardiovascular homeostasis also occur with fluctuations in the resistance of peripheral blood vessels as vascular beds regulate local blood flow to match supply with demand. These fluctuations in peripheral resistance may perturb central blood pressure and through the baroreceptor reflex, may also lead to a compensatory variation in heart rate.
Many types of medical instruments exist for studying heart rate variability. The instantaneous rate-meter is perhaps the earliest such instrument. This meter measures each RR interval through analog or digital circuitry and displays the instantaneous heart rate.
An improvement in the rate-meter is achieved by performing first order statistical evaluation on the RR-intervals. With mini- and micro-computer systems histogram displays of RR-interval differences may be generated along with their mean and standard deviations.
Another technique for heart rate variability analysis involves the study of spectral content of the instantaneous heart rate time series. In one approach to spectral analysis in animals, the computations are done on a computer. Akselrod, et al., Science, 213, 220-222 (1981) Hyndman, et al., Automedica, 1, 239-252 (1975). Such systems analyze data recorded on magnetic or punched tape. However, not only do these systems introduce additional errors during the recording process, they do not perform in real time. Furthermore, these systems are not multichannel in nature.
A Sparse Discrete Fourier Transform algorithm which may be implemented on a personal computer (CBM 2016) and which may perform on-line monitoring of heart rate variability, based on a low pass filtered cardiac event series is disclosed in Rompelman, et al., IEEE Trans. Biomed. Engineering, BME-29, 503-510 (1982). A specialized hardware device also exists for low pass filtering the cardiac event series by a stepwise convolution to create the low pass filtered cardiac event series. Coenen, et al., Medical and Biological Engineering and Computing, 15, 423-430 (1977). Nevertheless, these instruments possess a limited band width and a limited frequency resolution capability.
There exists a need for an instrument which provides multi-channel spectral analysis of an instantaneous heart rate and of a respiratory activity time series. There also exists a need for an instrument wherein such calculations are performed in real time at the bedside.
SUMMARY OF THE INVENTION
The present invention provides a method for diagnosis of malfunctions of the cardiovascular control system in a patient. A power spectrum of heart rate fluctuations in the patient is monitored, and a level below 0.1 (beats/min.).sup.2 is identified in the heart rate fluctuation power spectrum at a frequency between 0.04 and 0.10 Hz as indicative of cardiovascular instability or of heart failure.
Another method according to the present invention diagnoses malfunctions of the cardiovascular control system in a patient. A power spectrum of heart rate fluctuations in the patient is monitored, and a marked increase to above about 10 (beats/min.).sup.2 in a peak in the heart rate fluctuation power spectrum is identified at a frequency between 0.04 to 1 Hz as indicative of cardiovascular instability resulting from cardiac tamponade or depleted intravascular volume, or hemorrhage, or chylothorax, or acute hypoxia or acute hypercarbia.
Yet another method according to the present invention diagnoses malfunctions of the cardiovascular control system in a patient. A power spectrum of heart rate fluctuations in a patient is monitored, and a ratio of the area under a peak in the heart rate fluctuation power spectrum at a frequency between about 0.04 and about 0.1 Hz to the area under a respiratory peak in the heart rate fluctuation power spectrum centered at the mean respiratory rate as having an absolute value less than 2.0 as indicative of cardiac instability.
Still another method according to the present invention diagnoses malfunctions of the cardiovascular control system in a patient by monitoring a power spectrum of heart rate fluctuations in the patient and by identifying a ratio of the area under a heart rate power spectrum peak at a frequency between about 0.04 and about 0.1 Hz to the area under a peak in the respiratory heart rate fluctuation power spectrum centered at the mean respiratory rate as having an absolute value greater than or about 50 as indicative of cardiovascular stress.
An apparatus according to the present invention monitors a heart rate power spectrum. Means for converting an analog heart rate signal to a form acceptable to a programmable electronic device are coupled to a programmable electronic device.

BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 illustrates low frequency, mid-frequency and high frequency in the power spectrum of heart rate fluctuations in a dog according to the prior art;
FIG. 2 illustrates aspects of the cardiovascular control system according to the prior art;
FIG. 3 is a block diagram of apparatus for heart rate fluctuation power spectral analysis according to the present invention;
FIG. 4 illustrates address buffers and address decoding a data acquisition device according to the present invention;
FIG. 5 illustrates components according to the present for interfacing an ECG apparatus with a personal computer according to the present invention;
FIG. 6 illustrates a digital to analog converter according to the present invention;
FIG. 7 illustrates a ECG trigger according to the present invention;
FIG. 8 illustrates a portable calibrator according to the present invention;
FIG. 9A-B are a flow chart for software applicable to an embodiment of the present invention on a IBM personal computer;
FIG. 10 illustrates a trend for a stable patient according to the present invention;
FIG. 11, illustrates a trend display for an unstable patient according to the present invention;
FIG. 12 is an illustration of an instantaneous heart rate according to the present invention;
FIG. 13 is an illustration of an instantaneous heart rate fluctuation spectrum of the sort obtainable from apparatus according to the present invention;
FIG. 14 is a heart rate fluctuation power spectrum according to the present invention of a stable patient;
FIG. 15 is a heart rate fluctuation power spectrum according to the invention of an unstable patient;
FIG. 16 depicts distributions in LFP data obtained according to the present invention for stable and for unstable patients;
FIG. 17 graphically depicts distributions of RFP data according to the present invention for stable and for unstable patients; and
FIG. 18 graphically depicts data for LFP/RFP ratios according to the present invention for stable and for unstable patients.

DETAILED DESCRIPTION
Power spectral methods may be used to analyze the frequency content of fluctuations in heart rate and other hemodynamic parameters. Hyndman, et al., Nature, 233, 339-341 (1971); Sayers, Ergonomics, 16, 17-32 (1973). Short term (i.e., on a time scale of seconds to minutes) fluctuations in these parameters are concentrated in three principal spectral peaks as illustrated for a canine model in FIG. 1. Akselrod, et al., supra. One peak is centered at the respiratory frequency; this peak shifts with changes in the respiratory rate. The second identifiable spectral peak, the mid-frequency peak, occurs typically between 0.1 and 0.15 Hz. The oscillations associated with this second peak occur at 6-9 cycles per minute, a considerably lower frequency than the respiratory frequency, and are related to the frequency response of the baroreceptor reflex. The third peak of the spectrum typically occurs in the frequency band of 0.04 to 0.10 Hz. This low frequency peak is related to thermoregulatory fluctuations in vasomotor tone.
In one approach- to the spectral analysis of heart rate, properties of the heart rate fluctuations in the conscious dog may be related to the activity of three cardiovascular control systems--the parasympathetic nervous system, the sympathetic nervous system and the renin-angiotensin system. Akselrod, et al., Science, 213, 220-223 (1981). This model is further elaborated in Akselrod, et al., "Hemodynamic Regulation: Investigation by Spectral Analysis" (In Press). Heart rate fluctuations occurring at frequencies above roughly 0.1 Hz are mediated solely by the parasympathetic system. Blockade of the renin-angiotensin system leads to a dramatic increase in the amplitude of the low frequency peak. The effects of an autonomic blockade also exist in humans and changes in body posture alter sympathetic-parasympathetic balance as measured by the heart rate power spectrum. Pomeranz, et al., Am. J. Physiol., 248, H151-H153 (1985).
A simple model of the short term cardiovascular control system is illustrated in FIG. 2. Akselrod, et al., supra. In this model, heart rate is directly modulated by the sympathetic and parasympathetic nervous systems. Through a variety of receptors both these systems sense, fluctuations in cardiovascular parameters including arterial and venous pressures, vascular volumes, and correlates of blood flow and oxygenation. The parasympathetic system may respond over a wide frequency range while the sympathetic system may only respond at relatively low frequencies below roughly 0.1 Hz.
A hypothesis was proposed in Akselrod, et al., Science, 213, 220-223 (1981), that fluctuations in vasomotor tone associated with the low frequency heart rate fluctuations are not solely related to thermoregulation but also reflect local adjustment to resistance in individual beds of blood vessels in order to match local blood flow to local metabolic demand. Such fluctuations in peripheral vasomotor tone lead to fluctuations in central blood pressures which are in turn sensed by pressoreceptors. Stimulation of these pressoreceptors occasions an autonomically mediated baroreceptor reflex, which leads to compensatory fluctuations in heart rate at the corresponding frequency. In addition, the renin-angiotensin hormonal system senses blood pressure fluctuations and, through the elaboration of a substance called angiotensin II, plays the role of the guardian of the overall peripheral vascular resistance. Blockade of the renin-angiotensin system by a converting enzyme inhibitor, may remove this damping influence and may permit increased fluctuations in blood pressure and increased compensatory fluctuations in heart rate in the low frequency regime.
The critically ill infant or child prior to, during, and after cardiac surgery at times exhibits marked changes in heart rate, blood pressure, and peripheral perfusion. These changes may be of no clinical consequence or they may indicate the existence of a major unrecognized pathology whose first outward manifestation may be sudden cardiac arrest. To be able to quantify cardiovascular regulatory reserve permits objective assessment of a patient's cardiovascular stability as well as their response to medical and surgical interventions intended to improve cardiovascular function.
Spectral analysis of tape recorded records of ECG and respiratory activity from patients with complex congenital heart diseases and myocarditis reveals peculiarities in low frequency heart rate fluctuations not seen in studies of healthy children and adults. In particular: (1) low levels of low frequency heart rate fluctuations ar noted for critically ill patients in congestive heart failure, which levels revert to normal after surgical or medical treatment and (2) a marked increase in low frequency heart rate fluctuations is observed in patients with otherwise undetected cardiac tamponade.
A transitional microprocessor-based monitoring instrument, which utilized a Z-80 microprocessor and a S-100 bus, was constructed along with a data acquisition system which interfaced the microprocessor with a Hewlett-Packard 78341 patient monitor
A prototype system is described in Jerome C. Tu, "Microprocessor System for Real-Time Spectral Analysis Physiological Signals," Master of Department of Electrical Engineering and Computer Sciences, Science Thesis, Massachusetts Institute of Technology (1984) which is hereby incorporated by reference herein. An electrocardiogram (ECG) was inputed into a the data acquisition system for this prototype system from a patient monitor.
In the data acquisition system the analog voltage signal of the ECG was applied to the input of a variable frequency voltage controlled oscillator in the data acquisition system. A counter coupled to the output of the VCO provided a digital representation of the voltage associated with the ECG peaks. The largest voltage peak, called the R voltage peak and associated in the ECG with ventricular contraction, was used to trigger a clock. Each R peak load the value of the clock into a holding register and restarted the clock. The value of the clock provided a measure of the heart rate as the inverse of the time between beats. (i.e., as the RR internal).
The regular respiratory signal of a patient on a ventilator employed to obtain a respiratory spectrum was similarly obtained through a VCO. The respiratory frequency had to be manually entered to establish a fixed window for computing the power in the heart rate power spectrum in the respiratory peak.
Every 256 seconds the digitized ECG RR intervals were inputed in the microprocessor from the data acquisition system. A smoothed heart rate "tachometer wave form" was created as follows: (1) The instantaneous heart rate time series was computed from the stored RR intervals; (2) A 1024 point time series of the instantaneous heart rate was computed from the stored instantaneous heart rate time series by sampling the latter at 4 Hz; and (3). The mean heart rate computed from the 1024-point time series of instantaneous heart rate was subtracted from the smoothed series resulting in a "tachometer waveform"; The heart rate power spectrum was computed from the heart rate "tachometer waveform" as follows: (1) A 1024-Point Fast Fourier Transform was computed using 1024 points of the tachometer cardiac tachometer waveform; and (2) The heart rate power spectrum was computed by squaring the absolute value of the previously calculated transform.
As new data was inputted into the computer's buffer, the results of the smoothed cardiac tachometer signal, power spectrum and integral of power spectrum were outputted onto a printer. Thus for every 256-second time interval, a spectral representation of the preceding 256 seconds of instantaneous heart rate data was exhibited.
From the above data, the area under the low frequency peak (LFP) between 0.04 and 0.1 Hz and the area under the respiratory frequency peak within a peak width window of 0.2 Hz were determined. Trend graphs of LFP, RFP, and LFP/RFP ratio were created. 256 second data segments were rejected if (1) the patient was not in sinus rhythm, (2) transients and/or artifact were present on the cardiac "tachometer wave form", and (3) the LFP/RFP ratios were greater than 2 standard deviations from the mean for the study period.
The practical problems associated with the prototype monitoring instrument included the extremely tedious calculations required for use of the prototype with free-breathing patients and the large amount of data (as much as 50%, in some instances) which had to be discarded due to presence of motion artifacts. These artifacts resulted from virtually any disturbance of the patient, even one so slight as holding the patient's hand. The prototype system had no capacity to identify or reject artifacts or to examine the data for dropped beats and premature triggers.
Upon reviewing clinical studies performed using the prototype, it was discovered that not only were attenuated low frequency heart rate fluctuations associated with a severely compromised regulatory reserve but also that the ratio of the power in the heart rate power spectrum at low frequency to the power at the respiratory frequency provided an even sharper discriminatory index between stable and critically ill patients as illustrated in FIG. 3. In addition it was noted that this ratio was markedly elevated in the setting of moderate to severe congestive heart failure, cardiac tamponade, and prior to the development of malignant ventricular arrthymias.
A low value for LFP/RFP (<2) which is sustained for greater than one hour or a value greater than or about 50 is associated with a clinical course characterized by cardiac arrest and/or profound hypotension. At times this ratio may be the only clinical indicator of cardiovascular instability. The LFP/RFP ratio provides a sensitive and specific index of cardiovascular instability and may provide a clinically important continuous non-invasive probe of cardiovascular stability.
In order to further examine the diagnostic value of the power spectrum of heart rate fluctuations and to overcome the difficulties with the prototype, a multipurpose microcomputer based system including data basing, instantaneous heart rate and respiratory activity spectral monitor was developed using a Hewlett Packard Series 200 Computer and Multiprogrammer as available from Hewlett-Packard. Advantages over the original design include: (1) error correcting routines which correct automatically for motion artifact and missed triggerings of the EKG, thus permitting a substantial increase (>30%) in available data; (2) automated trending of spectral densities along with the instantaneous heart rate and respiratory activity time series; and (3) a data basing program which permits accurate temporal correlation of spectral densities with virtually every clinical intervention, routine ventilatory changes, hemodynamic, fluid monitoring and laboratory results. Software incorporating these advantages is included herein as Appendix A.
In a further improvement, programs and a data acquisition system and programs were developed for use with an IBM PC or compatible personal computer, This improvement is illustrated in FIGS. 3 through 12.
In FIG. 3, a block diagram of apparatus according to the present invention is illustrated. In FIG. 3, a source of an ECG signal 2 and a source of an electroplythsmogram signal 3 are contained within the patient monitor 4. A patient monitor for use with the present invention may be the system 2 infant monitor available from ARVEE, Incorporated, Battle Creek, Mich. Source 2 is connected to an ECG trigger 5 which is in turn connected to a personal computer 7. Source 3 is connected to an analog to digital interface 6. Interface 6 is connected to analog converter 8 which is connected in turn to a personal computer 7. Personal computer 7 receives input from and provides output to interface 6. Personal computer 7 is connected to a display 9.
Source 2 receives input from pregelled electrodes adhered to the chest wall and thigh of the patient. Source senses respiratory activity through a pair of electrodes by the impedance method. Personal computer 7 and display 9 are available as an IBM and a compatible display available from IBM, Incorporated, Armonk, N.Y. Elements 5, 6 and 8 are described below.
In a data acquisition device according to the present invention, address, buffers and address decoding, as illustrated in FIG. 4, receive input from a PC bus 10. Nodes 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25 and 26 are respectively connected to address lines A0, A1, A2, A3, A4, A5, A6, A7, A8, A9, A10, A11, A12, A13, A14 and A15 in PC bus 10. A first address buffer 100 has address inputs A0, A1, A2, A3, A4, A5, A6 and A7 which are respectively connected to nodes 11-18. Buffer 100 also has two gate inputs, 1G and 2G, which are connected to ground along with a ground output GND of buffer 100. A power supply input V.sub.CC of buffer 100 is connected to a node 102 at a potential of +5 volts.
A second address buffer 110 has address inputs A8, A9, A10, A11, A12, A13, A14 and A15 which are respectively connected to nodes 19-26. Buffer 110 also has two gate inputs, 1G and 2G, which are connected by way of a node 111 to ground. A ground GND output of buffer 110 is also connected to a common potential. Buffer 110 has a power supply input V.sub.CC which is connected to a node 112 at a potential of +5 volts.
A status buffer 120 has address inputs A16, A17, A18 and A19 which are respectively connected to nodes 27, 28, 29 and 30. Nodes 27-30 are respectively connected to an address enable line AEN, a reset line RES, an input/output read line IOR and an input/output write line IOW in PC bus 10. Buffer 120 has two gate inputs, 1G and 2G, which are connected by way of a node 121 to ground. A ground output GND of buffer 120 is also connected to ground by way of node 121. A power supply input V.sub.CC of buffer 120 is connected to a node 122 at a potential of +5 volts.
According to the present invention, a data acquisition system board which is both reliable and compatible with a personal computer (PC) bus, preferably adheres to the timing requirements and the loading requirements supplied by the PC bus. This means that all connections to the PC bus should be buffered so that the load provided at any input or output of the bus is equivalent to 1 LS TTL load and high speed CMOS integrated circuits are fed for this purpose.
Because there are multiple devices attached to the address bus, address buffers are provided. This is done by buffers 100 and 110. Parts used for buffers 100, 110 and 120 are normally gated, but the gate enables, 1G and 2G, are tied to ground so that the gates are always enabled. Some of the status lines on the PC bus are buffered by a chip 120, in particular: the reset line RES; the read and write lines IOR and IOW respectively, for the input/output (IO) channels; and the address enable AEN.
An address decoder according to the present invention, as illustrated in FIG. 4, includes a three to eight line decoder 130. Decoder 130 has three line inputs A, B and C which are respectively connected to outputs B2, B3 and B4 of buffer 100. Decoder 130 has gate inputs G2A and G2B which are respectively connected to outputs B5 and B6 of buffer 100. A power supply VCC input of decoder 130 is connected to a node 131 at a potential of +5 volts while a ground GND output of decoder 130 is connected to a common potential. Outputs Y0, Y1, Y2, Y3, Y4, Y5, Y6 and Y7 are connected to inputs of a NAND gate 140.
A NAND gate 151 has an input connected to each of outputs B8, B9 and B10 of buffer 110. An output B10 of buffer 110 is connected to an input of an inverter 152 which has an output connected to an input of NAND gate 151. Similarly, outputs B12, B13, B14 and B15 of buffer 110 are respectively connected to an input of each of inverters 153, 154, 155 and 156, each of which has an output connected to an input of NAND gate 151. NAND gate 151 has an output connected to an input of an inverter 157.
A NAND gate 158 has an input connected to an output of inverter 157 and has an output connected to an input of an inverter 159. An inverter 160 has an input connected to an output B7 of buffer 100 and has an output connected to an input of NAND gate 158. Likewise, an inverter 161 has an input connected to an output B16 of buffer 120 and has an output connected to an input of NAND 158. An output of inverter 159 is connected to a gate input G1 of decoder 130.
So that devices on the board are recognized at a particular IO channel address, address decoding is provided. In this particular case, a fixed address location, location hex 700 to 71F (a total of 32 channels), is used. The decoding of the fixed upper bytes in the address is provided by a combination of nine inverting gates, 152, 153, 154, 155, 156, 157, 159, 160 and 161, and NAND gates 151 and 158. These elements, in combination with decoder 140, provide chip enable signals which can be used to select one or another of the functional chips on our board. Each of the eight chip enable signals correspond to a block of four channels. For example, a chip select #0 from output to of decoder 130 corresponds to channels hex 700, 701, 702 and 703.
A logic network for driving a data buffer, as illustrated in FIG. 4, includes a NAND gate 171, an inverter 172 and a NAND gate 173. An output of inverter 172 is connected to a first input of NAND gate 173 while an output of NAND gate 140 is connected by way of a node 174 to a second input of NAND 173 and to a first input of a NAND gate 175. A second input of NAND gate 175 is connected to an output of NAND gate 171.
In addition, a node 181 is connected to an output B0 of buffer 100. A node 182 is connected to an output B1 of buffer 100. Nodes 183 and 184 are respectively connected to output Y0 and output Y7 of decoder 130. Nodes 185 and 186 are respectively connected to an output of NAND gate 175 and an output of NAND gate 173. A node 187 is connected to an output B17 of buffer 120. A node 188 is connected an output B18 of buffer 120, to a first input of NAND gate 171 and to an input of inverter 172. A node 189 is connected to a second input of NAND 171 and to an output B19 of buffer 120.
Additional chips are used to provide logic which drives a data buffer connected to a data bus. The data bus is bidirectional in order to both transmit data to and from devices on the board. In order that this be accomplished, one must determine at any time whether or not data is either being read from or written to the board. This logic is supplied by NAND gate 171, NAND gate 173, AND gate 175 and inverter 172 which translates the read and write signals for the input/output (IO) channel into an output enable and a transmit enable for a data buffer. The apparatus of FIG. 5 may be used to properly interface a device to the PC bus 10.
As illustrated in FIG. 5, components according to the present invention for interfacing an ECG apparatus with a personal computer include a port expander 200. Port expander 200 has four sets of 8 nodes each, the four sets correspond to four ports A, B, C and D. The outputs for port A are A0, A1, A2, A3, A4, A5, A6 and A7. The inputs corresponding to port B are B0, B1, B2, B3, B4, B5, B6 and B7. Outputs corresponding to port C are C0, C1, C2, C3, C4, C5, C6 and C7. A set of outputs corresponding to port D includes D0, D1, D2, D3, D4, D5, D6 and D7. Expander 200 has a chip select input CS connected to node 184. Expander 200 also has a read input RD and a write input WR respectively connected to nodes 188 and 189. Expander 200 has two address inputs, AD0 and ADl which are respectively connected to nodes 181 and 182. A reset RES input of expander 200 is connected to node 187. Inputs A0, A1, A2, A3, A4, A5, A6, A7 are respectively connected to nodes 291, 292, 293, 294, 295, 296, 297 and 298. Outputs D0-D7 are respectively connected to nodes 208, 207, 206, 205, 204, 203, 202 and 201 which define a data bus. A power supply input V.sub.CC of expander 200 is connected to a node 209 at a potential of +5 volts. A ground GND output of expander 200 is connected to a common potential.
Port expander 200 is used to overcome the low speed of the data bus on both A/D converter 260 and a digital analog converter. This permits slowing down the read and write signals inasmuch as they may be provided artificially on port C of expander 200 or as chip select signals from address decoder 130. Port C of expander 200 is a bit addressable register which allows one to individually select or deselect bits without affecting any of the other bits. This is accomplished by sending a one byte command to expander 200. Because expander 200 is given the control function, the address of expander 200 is the highest address in the set of channels. In other words, expander 200 occupies IO channels hex 71C to hex 71F. The ports A, B and C on expander 200 are addresses 71C, 71D and 71E, respectively, and the control register internal for expander is at input/output I/0 channel 71F.
A timer 220 according to the present invention has two address inputs, AD0 and AD1 respectively connected to nodes 181 and 182. Timer 220 also has a read input RD connected to node 188, a write input WR connected to node 189 and a chip select input CS connected to node 184. A first gate input G0 is connected to the C0 of expander 200 while a second gate input G1 and a third gate input G2 are both connected by way of a node 223 to output C1 of expander 200. Timer 220 has three clock inputs CLK0, CLK1 and CLK2, of which CLK1 is connected by way of node 222 to an output OUT0 of timer 220 and input CLK2 is connected to an output OUT1 of timer 221 by way of a node 31. An interrupt request line IRQ4 within PC bus 10 is also connected to node 31.
An output OUT2 is connected to a non-inverting input of an operational amplifier 224, an inverting input and an output of which is connected to a node 400.
A power supply input V.sub.CC of timer 220 is connected to a node 221 which at a potential of +5 volts.
Timer 221 has seven outputs D0, D1, D2, D3, D4, D5, D6 and D7 which are respectively connected to nodes 208, 207, 206, 205, 204, 203, 202 and 201. A ground output of timer 220 is connected to a common potential.
Timer 220 includes three 16 bit timers which are addressed at hex locations 704, 705, 706, and 707. In other words, they are provided by chip select 1. The three clocks on timer 220 are connected in series which effectively converts it into a 48 bit counter. However, in the operation of the program, some of the bits in this counter are thrown away because the reset values are less than 65,536. The three clock registers are used in the following way. Counter 0, corresponding to input CLK 0, counts an onboard time base to be discussed later and provides an output which gives the minimum resolution of the heart rate counting. In other words, it provides the counter time base for measuring the heart rate. Counter #1, corresponding to input CLK 1, counts the heart rate counter time base and provides as an output an interrupt at IRQ4. This signal drives the sampling of the respiratory signal at a constant frequency, and is also used to measure interbeat intervals. In the standard data collecting mode, where one is interested in measuring the respiratory signal at 4 hertz intervals, this means that the counter 0 is set to generate output pulses at 11 microsec. intervals and that these pulses are in turn counted by counter 1 to generate 4 hertz pulses which are used to drive data acquisition from the respiratory signal. The last counter register, counter #2, corresponding to input CLK2, is used to count the number of respiratory sampling pulses which have been supplied. This functions as an overflow counter and always has the reset value of 65,536. Thus the counter measuring interbeat intervals effectively overflows only every 65,536 respiratory sampling times, which is far in excess of what would be required to recover dropped beats which occur because the heart rate is not adequately detected.
A counter 240 has an input 1A connected to a clock line PC CLK in PC bus 10 by way of a node 32. Counter 240 has a first output 1QA connected to the CLK0 input of timer 220. Counter 240 has a second output 1QB and has a third output 1QC. A clear input CLR1 of counter 240 and a ground output GND of timer 240 are connected to a common potential by way of a node 242.
A data output buffer 280 has an output enable input OE connected to node 185 and has a transfer enable input TE connected to a node 186. Eight data inputs, A0, A1, A2, A3, A4, A5, A6 and A7, of buffer 280 are respectively connected to nodes 208, 207, 206, 205, 204, 203, 202 and 201. A power supply VCC input of buffer 280 is connected to a source of potential at +5 volts. A ground GND output of buffer 280 is connected to a common potential. Outputs B0, B1, B2, B3, B4, B5, B6 and B7 of buffer 280 are respectively connected to data lines in PC bus 10 by way of nodes 33, 34, 35, 36, 37, 38, 39 and 40.
The time base for this clock system is provided by counter 240. Timer 220 counts only at a rate of 2.6 MHz megahertz which is exceeded by the IBM PC bus clock of 4.77 megahertz. The IBM PC bus clock is divided by 2 using counter 240 and the result used to provide a time base at 2.38 megahertz for timer 220. The 4.77 megahertz clock is also divided by 8 to provide a 596 kilohertz clock which is used to drive an analog to digital (A/D) converter. A/D converter 260 uses this clock signal in order to properly execute the successive approximation scheme to convert analog inputs into digital outputs.
A/D converter 260 has an output enable input 0E connected to output C4 of expander 200. A/D converter 260 also has three inputs A, B and C which are respectively connected to outputs C5, C6 and C7 of expander 200. A clock input CLK of A/D converter 260 is connected to the 1QC output of counter 240. An address latch enable ALE and a start input STR of A/D converter 260 are connected to a node 261. A power supply V.sub.CC input and a reference voltage +V.sub.REF input of A/D converter 260 are connected to a node 262 at a potential of +5 volts. A reference voltage -V.sub.REF output and a ground GND output of A/D. converter 260 are connected to a common potential by way of a node 263. A/D converter 260 has seven outputs D0, D1, D2, D3, D4, D5, D6 and D7 which are respectively connected to inputs B0, B1, B2, B3, B4, B5, B6 and B7 of expander 200. In addition, A/D converter 260 has an end of count EOC output connected to a first input of the NAND gate 264, an output of which is connected to an input of an inverter 265. A second input of NAND gate 264 is connected to an output of an inverter 266 which has an input connected to node 187. An output of inverter 265 is connected to node 261.
A/D converter 260 has a signal input IN connected to a node 267. An output of an operational amplifier 268 is connected to node 267 and to a first lead of a resistor 269. A second lead of resistor 269 is connected to a first lead of resistor 270, a second lead of which is connected to a source of potential at -5 volts. The first end of resistor 270 is also connected to an inverting input of amplifier 268 and to a first end of a resistor 271. A non-inverting input of amplifier 268 is tied to ground. A second end of resistor 271 is connected to a node 272 which provides an analog signal input ANA IN for the apparatus according to the present invention.
A/D converter 260 is connected to port B of port expander 200. This A/D has built into it its own 8 channel analog multiplexer which allows the selection of one of eight analog signals to, be converted. The channel select corresponding to inputs A, B and C of converter 260 is connected to port C on bytes 5, 6 and 7.
Because A/D converter 260 operates from 0 to 5 volts, analog input at input IN should be in the range of 0 to 5 volts or an input buffer should be supplied to alter this input range. However, in keeping with general practices for safety and isolation, input IN should always be provided with an analog buffer to provide isolation for both the computer and the instrument being monitored. As illustrated, the input buffer is provided by operational amplifier 268. This amplifier converts a bipolar analog input of plus or minus 5 volts to a single unipolar input of 0 to 5 volts at input IN. This analog input is used to monitor the respiration.
A/D converter 260 is set up in a free running mode such that it continuously does conversions on the analog signal. The end-of-conversion pulse at output EOC is used to generate a start pulse for the A/D so that as soon as an end of conversion occurs it a new conversion is started. This is the reason for the two gates connected between end of conversion output EOC and the start input STR. In order to prevent latchup of the device on power up, the reset line at node 187 is also used to generate a start pulse. This means that the device will always function even after being powered up. Also, in order to update A/D converter 260 as frequently as possible, the address latch enable ALE, which is used to latch in the address value for the channel to be monitored, is re-latched at every start pulse.
As illustrated in FIG. 6, digital analog (D/A) converter 300 has inputs D0, D1, D2, D3, D4, D5, D6 and D7 which are respectively connected to nodes 298, 297, 296, 295, 294, 293, 292 and 291 as illustrated in FIG. 4. Converter 300 has a write WR input connected to node 183 and has a feedback input RFB. Converter 300 also has a power supply V.sub.CC input, a reference voltage V.sub.REF input and an input latch enable input ILE all of which are connected to a source of potential at +5 volts by way of a node 301. Converter 300 has an analog ground AGND and a digital ground output DGND, both of which are connected by way of a node 302 to a common potential.
Converter 300 has a first output OUT1 and a second output OUT2 which are respectively connected to an inverting and a non-inverting input of an operational amplifier 303. The non-inverting input of amplifier 303 is also connected to a common potential by way of a node 305. Amplifier 303 has an input connected to a node 306 at a potential of +12 volts and an input connected to a node 307 at a potential of -12 volts. An output of amplifier 303 is connected to a node 308 which is connected to the RFB input of converter 300 and to a first end of a variable resistor 309. A second lead of variable resistor 309 is connected to a first lead of a variable resistor 310, a second lead of which is connected to a node 311 at a potential of +5 volts. The second lead of resistor 309 is also connected to an inverting input of operational amplifier 312 and to a first lead of a resistor 313. A non-inverting input of amplifier 312 is connected to ground. A second lead of resistor 313 is connected to an output of amplifier 312 and to a node 391 which serves as an analog output for the apparatus according to the present invention.
Port A of expander which is at location 71C, is attached to a D/A converter data bus which, includes nodes 291-298. The write latch signal for the D/A converter is provided by chip select #0. In other words, any dummy byte written to any of the addresses 700, 701, 702 or 703 hex will cause a write pulse to be sent to D/A converter 300, thereby latching the data on port A of expander 200 into the D/A converter 300 and allowing an analog signal to be generated corresponding to the digital input. The output of D/A converter 300 chip is in the form of differential currents generated at outputs OUT1 and OUT2. A system having two operational amplifiers is employed to convert these currents to a voltage. Amplifier 303 is a differential current to voltage converter which provides a signal from 0 to 5 volts. Amplifier 312 converts the signal to a bipolar plus or minus 5 volt signal. Feedback control for the current to voltage converter is provided in D/A converter 300 through input RFB so that in actuality three connections are made from the D/A chip to the first operational amplifier. Because the D/A converter is an 8 bit device, this provides 256 voltage levels which are linearly distributed between plus and minus 5 volts. This D/A output may be used to generate calibrating signals or other control signals.
As illustrated in FIG. 7, a source of an ECG signal is connected by way of a node 400 to a non-inverting input of an operational amplifier 401 in an ECG trigger 60. An input of amplifier 401 is connected to a node 402 at a potential of plus 12 volts. An inverting input of amplifier 401 is connected to an output of amplifier 401 and to a non-inverting input of an operational amplifier 406. A first lead of each of resistors 403a, 403b, 403c, 403d, 403e, 403f, 403g, 403h and 403i is connected to the output of amplifier 401 while the second lead of resistor 403i is permanently connected and a second lead of one other of resistors 403a through h is connected to a node 410 by a jumper. A first lead of capacitor 404 is connected to node 410 while a second lead of capacitor 404 is connected to a node 405 at a potential of minus 12 volts. An inverting input of amplifier 406 is connected to a cathode of a diode 407, an anode of which is connected to an output of amplifier 406. The cathode of diode 407 is also connected to a first lead of capacitor 408 and a first lead of each of resistors 410a, 410b, 410c, 10i q and 410e, the second lead of resistor 410e is permanently connected and the second lead of one other of which is connected to a node 410 by a jumper 411g. A non-inverting input of an operational amplifier 412 is also connected to the cathode of diode 407 while an inverting input of amplifier 412 is connected to the output of amplifier 406. An input of amplifier 412 is connected to a node 413 at a potential of minus 12 volts. A first lead of resistor 414 is connected to the output of amplifier 412 while a second lead of resistor 414 is connected to a cathode of a diode 415 an anode of which is connected to ground. The cathode of diode 415 is also connected to an input of a Schmitt trigger 416 an output of which is connected to a line designated IRQ3 in PC bus 10 by way of a node 491.
ECG trigger 60 has an input buffer consisting of a non-inverting buffer of an amplifier 401 which isolates the ECG signal from the rest of the board. As illustrated in FIG. 5, the EKG trigger functions in the following manner. The R wave, which is larger than any other signal in the ECG, causes capacitor 408 to charge up to a certain value corresponding to the peak of the R wave. Any values beneath the peak of the R wave will be rejected by amplifier 403 so that no output occurs. Between R waves, the voltage on capacitor 405 decays slowly with a rate given by the RC time constant of capacitor 405 and the resistance across elements 410a-f. The voltage on the capacitor is sent to the inverting input on amplifier 403 and is used as a threshold for the R wave of the EKG. Therefore, as the electrocardiogram is being passed to the non-inverting input of amplifier 406, the only time that the operational amplifier has a positive output is when the EKG signal is larger than the voltage on capacitor 405. Whenever this occurs, capacitor 408 is immediately charged up to the value at the EKG input. In other words, the voltage on capacitor 408 is a sort of envelope on the top of the electrocardiogram, although its decay rate is limited by the RC time constant. Diode 407 insures that the envelope function which is provided by capacitor 408 is the upper envelope and not the lower envelope. The lower envelope is provided by reversing the polarity of diode 407.
The RC network of capacitor 405 and resistors 403a-i provides a low pass filtered ECG. The voltage on capacitor 405 is the baseline for the ECG, which may vary. The array of jumper selected resistors 410a-e allows variation of the time constant of the RC network containing resistors 406a-e and capacitor 408. Thus, this latter network which monitors the ECG envelope is referenced to the ECG baseline present on capacitor 404 permitting accurate tracking of the envelope and therefore better R wave detection. As a further improvement, the jumpers may be replaced with analog switches controlled by the personal computer in order to give the computer control of RC time constant selection.
An output from ECG trigger 60 is generated by connecting amplifier 412 in parallel with peak detector amplifier 406 so that the inputs are reversed. The result is that the output polarity is inverted. Because the amplifiers 401, 406 and 412 are operating from a plus 12 volts to minus 12 volts supply, but the logic levels on the board are only from 0-5 volts, resistor 414 and a diode 415 are used to clamp the output value of the amplifier 412 between 0 and 12 volts. This signal is then passed to a Schmitt trigger 416, which is a single conditioning device. The output of this signal conditioner is finally provided to PC bus 10 in order to drive interrupts at interrupt request 3 (IRQ3) indicating the currents of a R wave. ECG trigger 60 may be modified to allow selection of various decay rates for the envelope and also to provide a floating threshold for the 0 point of the EKG. The ECG triggers if the R wave passes above 0 volts. However, it can be imagined that sometimes the baseline will drift far enough below 0 volts that the R wave does not cross 0 volts and in such a case this trigger would never detect the R wave. This is corrected by connecting the second leads of the charging capacitor 408 and on the selected discharging resistor of 406a-f may be connected to a low pass filter consisting of a capacitor 405 and a selected one of resistors 403a-f (to choose various discharge rates) which low pass filters the electrocardiograms and essentially selects out the baseline. This means that instead of measuring the R wave with respect to 0 volts, the R wave may be measured with respect to the floating baseline of the electrocardiogram. The jumper selected resistor selects an RC time constant much greater than the RR interval. So long as the baseline does not drift faster than one R wave in approximately 10 heart beats, this means that this trigger will successfully detect all R waves. Selecting one of resistors 410a-f allows variation of the RC time constant of elements 408 and 410a-f.
As illustrated in FIG. 8, in a portable calibrator 70 according to the present invention, an operational amplifier 500 has a non-inverting input connected to a first lead of each of resistors 501, 502 and 503. A second lead of resistor 501 is connected by way of a node 503a to a positive voltage source while a second lead of resistor 502 is connected by way of a node 504 to a negative voltage source. An inverting input of amplifier 500 is connected to a first lead of a capacitor 505, a second lead of which is connected by way of a node 506 to a negative voltage source. The inverting input of amplifier 505 is also connected to a first lead of a variable resistor 507 and to a first lead of a resistor 508 a second lead of which is connected to an output of amplifier 500. The output of amplifier 500 is also connected to a second lead of resistor 503. Amplifier 500 has an input connected by way of a node 509 to a positive voltage source and by way of a node 510 to a negative voltage source.
A second lead of resistor 507 is connected to a non-inverting input of an amplifier 511, an inverting input of which is connected to an output of amplifier 511 by way of a node 591 which provides an output port for a simulated respiratory frequency.
A first lead of a resistor 512 is connected to node 591 while a second lead of resistor 512 is connected to a first lead of a resistor 513 and to a first lead of a capacitor 514, a second lead of which is connected by way of a node 515 to a negative voltage source. A second lead of resistor 513 is connected to an output of an operational amplifier 514 and to an inverting input of amplifier 515 is connected to a first lead of a resistor 516, to a first lead of a capacitor 517 and to an inverting input of an operational amplifier 518. The second lead of capacitor 517 is connected by way of a node 519 to a negative voltage source. A non-inverting input of amplifier 518 is connected to a first lead of each of resistors 520, 521 and 522. A second lead of resistor 520 is connected by way of a node 523 to a positive voltage source while a second lead of resistor 521 is connected by way of a node 524 to a negative voltage source. A second lead of resistor 522 is connected to an output of amplifier 518 and to a second lead of resistor 516.
An inverting input of an operational amplifier 525 is connected to the first lead of resistor 513 and to a first lead of a variable resistor 526. A non-inverting input of amplifier 525 is connected to a first lead of each of resistors 527, 528 and 529. A second lead of resistor 527 is connected to a node 530 at a positive potential while a second lead of resistor 528 is connected by way of a node 531 to a negative voltage source. A second lead of resistor 529 is connected to a second lead of resistor 526 and to an output of amplifier 525 at a node 592 which provides a square wave output simulating a modulated heart rate pulse. A first lead of a capacitor 532 is connected to node 592 while a second lead of capacitor 532 is connected by way of a node 593 to a first lead of a resistor 533, a second lead of which is connected to ground. Node 593 provides an output port for a spike output simulating the R wave of an EKG.
The source of positive potential for the portable calibrator 70 may be at a voltage between about plus 5 and about plus 18 volts. Similarly, the negative voltage source for portable calibrator 70 may be at a potential of about minus 18 volts to about minus 5 volts.
Portable calibrator 70 provides test signal for the heart rate spectral analysis hardware which, although not of a truly calibrated nature, does allow one to evaluate whether or not the software and hardware is functional. Each of the output signals provided is a triangle wave which represents the respiration and a frequency modulated pulse train representing the heart rate. The modulation of the heart rate is provided at two frequencies which simulate a respiratory modulation and also a low frequency modulation.
The basic circuit of calibrator 70 for providing each pulse train consists of an oscillator having one operational amplifier as typified by the respiratory frequency modulator. A charging capacitor 505 and a variable resistor 507, provide an RC circuit which is charged by the output of the amplifier 500. It is also discharged by the amplifier 500 when the output of the amplifier 500 is low. Progressive cycles of the oscillator consist of charging and discharging the capacitor at the rate prescribed by the RC circuit. The reference level which determines whether or not one is discharging or charging is provided at the non-inverting input of the amplifier 500.
Suppose, for example, that capacitor 505 begins as being completely discharged, then the voltage at the inverting input for the operational amplifier 500 is low. The output of the operational amplifier 500 is therefore high and this means that the input at the non-inverting input is 2/3 the voltage between the negative voltage source V and the positive voltage source V+. Thus the capacitor 505 begins to charge. When the capacitor voltage exceeds the threshold at the non-inverting input of the operational amplifier 500, the output of operational amplifier 500 changes sign and capacitor 505 begins to discharge. However, when the output of the amplifier 500 changes to the negative side, then the threshold voltage at the non-inverting input is changed and now becomes only 1/3 the way from the negative voltage source to the positive voltage source. This means that the voltage on the charging capacitor 505 varies between 1/3 and 2/3 the difference between the negative and the positive voltage source. This determines the range of output on capacitor 505. The voltage at capacitor 505 is buffered by a non-inverting buffer 511 and this provides the respiratory signal at node 591.
An identical oscillator is used to provide low frequency modulation. The difference in the two frequencies is obtained by adjusting the respective variable resistors, 505 and 517, which set the RC time constants. The outputs of these two modulators are fed by resistors 512 and 513 into the charging capacitor 514 for the heart rate.
The heart rate oscillator is similar in design and consists of variable resistor 526 and capacitor 532 which charges and discharges in cycles with the range of voltages on the capacitor ranging between 1/3 the distance from the negative voltage source to the positive voltage source to 2/3 the voltage between the negative voltage source and the positive voltage source. Resistors 512 and 513, which connect the outputs of the low frequency and respiratory frequency modulators to the heart rate modulator, allow a small amount of current to flow into charging capacitor 514 of the heart rate modulator. This alters the charging rate of capacitor 514 and thereby affects the rate at which the heart rate oscillator oscillates. For example, on a positive cycle of the respiratory frequency modulator, the heart rate capacitor is charging more rapidly towards the plus side because more current is being supplied on the plus side of the cycle. Finally, the output of the heart rate modulator is sent through an RC filter comprising capacitor 532 and resistor 533 which converts the square wave output of the heart rate modulator into a spike output which may be sent to an R wave detector. Notice that the spike output includes both positive and negative spikes so that an R detector which depends on a high frequency filtering function may be discharging at twice the heart rate, inasmuch as it may trigger on both positive and negative spikes.
As illustrated by a block diagram in FIG. 9, a block diagram may be constructed for the main program (designated Syncts19) and for sub-routine modules (Sync7s, Gwindow3, and Fgraph8). This block diagram may be used in order to better interpret a complete program for heart rate fluctuation spectral analysis useful on an IBM personal computer, as illustrated in Appendix B. Although programs are provided for a Hewlett-Packard and an IBM computer herein, the software and other aspects of the present invention may be readily modified for use with other mini- and micro-computers.
In the program of Appendix B is a routine for removing artifacts from a detected heart rate provided for by an electrocardiograph machine. This program computes histograms from the heart rate data in order to generate a tachometer waveform. The most common rate on the histogram is selected as the correct rate and other rates are interpreted in light of it. Specifically, in order to correct for a spurious extra trigger, where a first and a second beat are close together while a third beat is spaced at an abnormally long interval, the second beat is discarded if the first beat to second beat interval is less than a predetermined value. The resulting interval between the first and the third beats is divided by an integer in order to provide a more normal interbeat interval. Where a trigger has been missed, so that a first and a second beat are separated by an interval which is approximately a multiple of a normal intrabeat interval, the intrabeat interval is divided by that multiple, most commonly two, in order to provide a more correct interval length. If the slewing rate of the heartbeat samples is outside of an acceptable range of slewing rates determined as a function of a mean variance, and the problem cannot be identified as a missed trigger or as a spurious extra trigger, or if the three previous intervals have been corrected, a determined mean interval, against which all other intervals are judged, is substituted for the inappropriate interval.
The slew rate is calculated on a moving average of the heart rate waveform and corrects for triggers that fall within the parameters of 0.05 Hz (3 beats/min.) per beat and five times the maximum slew. This artifact-correcting routine never slews more than 10 percent of the heart rate waveform.
Within the software of Appendix A is a graphic routine for trending heart rate fluctuation spectral data. The parameters of LFP, RFP, LFP/RFP ratio and heart rate are plotted on a graph over time to show trends in the four parameters. These trends may then be studied in order to examine the effects of various clinical interventions. Values for the parameters heart rate, LFP/RFP ratio, LFP and RFP are stored and may be called up at any point in time through a graphing routine in order to provide a graphic depiction of the course of a patient's condition. This sort of graphic depiction is illustrated for a stable patient in FIG. 10 and for an unstable patient in FIG. 11.
Also present in the program of Appendix B, a routine is provided for the segmentation of data and subsequent reanalysis. In this routine, data from the analog to digital converter 260 is collected continuously into a buffer and is dumped to a disk in blocks of 1,024 numbers (2,048 bytes equals 1,024 words and each block is referred to as a record or EPOCH). The time of heartbeat occurrence as measured by the signal provided by outputs OUT1 and OUT2 of timer 220 are collected continuously into two buffers (hb buffer 1 and hb buffer 2). These times are dumped to the disk in blocks of 1,024 pairs of numbers (1,024 from each buffer which equals 2,048 bytes or 1,024 words each). Because the heart rate is less than the sample rate of A/D converter 260 as required by signal processing, there are fewer heartbeat disk dumps.
In order to properly analyze data, the A/D and heartbeat data must correspond to the same time interval for the purpose of doing correlations. The correspondence may be determined from (1) the record number in a A/D file and (2) the absolute of the times stored in the heartbeat file (time differences used for intrabeat intervals). The instantaneous heart rate signal is generated backwards in time from the heartbeat corresponding to the last A/D sample in the record of interest. This means that if the heart rate signal is analyzed on a frequency scale not corresponding to the respiration data (e.g. respiration sample at 16 Hz but a heart rate analysis at 0 to 4 Hz) then the heart rate waveform extends backwards in time beyond the beginning of the present A/D record. This means that the heart rate waveform overlaps the heart rate waveform corresponding to the previous A/D records.
Overlapping permits lower frequency analysis than would be possible if only data corresponding to the present record were used (as in the prototype apparatus). Also, overlapping leads to the smoothing of parameters and to the subsequent reduction of fluctuating artifacts. In addition, it becomes less critical at what point analysis begins.
A calibration program providing a software driven calibrator, which may provide more realistic spectral data than the portable calibrator of FIG. 8, is contained within the program of Appendix A for a Hewlett-Packard micro-computer. FIG. 12 is a program which, although not tested, is believed to provide the same sort of software-driven calibration for an IBM personal computer through the data acquisition system of FIGS. 4 through 7.
In general, outputs OUT0 and OUTl of timer 220 in FIG. 5 generate a time base used via interrupt request line IRQ4 to clock data from a buffer to D/A converter 300. This buffer contains a respiratory waveform which may be a sign wave or any selected waveform as obtained by changing the contents of the buffer. Output OUT2 of timer 220 generates a heartbeat pulse as its output. In order to work properly, this pulse must be returned to the ECG trigger through node 400 or directly to interrupt request line IRQ3. If the latter course is chosen, however, node 491 must be disconnected from the output of Schmitt trigger 416. By returning the pulse to the ECG trigger, the computer is informed that the timer is through counting the present RR interval and needs a new interval to be loaded into a timer register of timer 220.
Through the use of the apparatus according to the present invention, a display of instantaneous heart rate as provided by an electrocardiograph machine, and as illustrated in FIG. 12, may be converted into an instantaneous heart rate fluctuation spectrum as illustrated in FIG. 13. A typical spectrum for a stable patient is illustrated in FIG. 14 while a typical spectrum for an unstable patient is illustrated in FIG. 15.
Example I and Example II relate respectively to diagnosis and to treatment employing the present invention.
Parts suitable for use in construction of the apparatus as illustrated in FIGS. 3 through 8 may include those as listed in Tables I, II, III and IV.
TABLE I______________________________________Element No. Part No. Manufacturer, Location______________________________________100, 110, 120 74HC244 National Semiconductor Santa Clara, California130 74HC138 National Semiconductor Santa Clara, California140, 151, 158 74HC30 National Semiconductor Santa Clara, California152, 153, 154 74HC04 National Semiconductor155, 156, 157 Santa Clara, California159, 160, 161172, 265, 266171, 173, 185 74HC00 National Semiconductor264 Santa Clara, California200 8255A-5 Intel Corporation Santa Clara, California220 8253-5 Intel Corporation Santa Clara, California224240 74HC393 National Semiconductor Santa Clara, California260 ADC0808 National Semiconductor Santa Clara, California268, 303, 312 LM324AN National Semiconductor401, 406, 412 Santa Clara, California500, 511, 515518, 525280 8286 Intel Corporation Santa Clara, California300 DAC0830 National Semiconductor Santa Clara, California416 74HC14 National Semiconductor Santa Clara, California______________________________________
TABLE II______________________________________Diodes Element Part No.______________________________________ 407, 415 IN4148______________________________________
TABLE III______________________________________ResistorsElement No. Value (in Ohms)______________________________________403i, 410 2.2k269 5k270, 271, 409 10k 403h 15k 403g 27k 403f 56k313 82k309, 310, 501, 100k (variable)502, 503, 520,521, 522, 527,528, 529, 533,403e403d, 410d 220k403c, 410e 560k508, 516, 526, 1 M (variable)403b, 410b512, 513, 2.2 M403a, 410a______________________________________
TABLE IV______________________________________CapacitorsElement No. Value (in microFarads)______________________________________405 2.2404, 505, 517 10532 0.1514 1______________________________________
EXAMPLE 1
Heart rate spectral analysis was applied to the study of congestive heart failure in infants and children. Congestive heart failure is characterized by a marked alteration in cardiovascular regulation. However, many cardiovascular functions which are normally monitored in cardiac intensive care units (such as: mean heart rate; arterial blood pressure; arterial blood gases; left arterial pressure and right arterial pressure; right atrial, left atrial and pulmonary artery oxygen saturations; the peripheral pulses; peripheral perfusion; and cardiac output) may not clearly indicate a critically unstable cardiovascular condition. The usually-monitored cardiovascular function parameters may be within a normal range immediately before a major cardiovascular crisis, such as hypotension or cardiac arrest, inasmuch as the cardiovascular regulatory system maintains these parameters within a normal range up to the point of system failure.
Twenty-nine infants and children were studied in a cardiac intensive unit. Of the twenty-nine patients, twenty-six have undergone a cardiac surgical procedure. The patients were studied for a minimum of three hours and a maximum of twenty-seven hours, with a mean study time of eight hours. EKG for cases were recorded and analyzed continuously in real time during the study time.
Data for a particular patient was analyzed only if the patient was in sinus rhythm. The patient's clinical course during the period of study was reviewed and, in particular, major events such as cardiac arrest, hemorrhage and profound hypotension were correlated with spectral analysis data. Administration of medication and the mode of ventilation were noted.
Real time heart rate spectral analysis was performed on a dedicated personal computer using a 6809E Motorola Microprocessor-Based System. A data acquisition system interfaced the computer with a patient monitor, available from Hewlett-Packard, Palo Alto, Calif., as Model No. 78341.
The heart rate power spectrum was calculated in continuous 256 second data epochs. A QRS synchronization pulse from the patient monitor was used to determine an RR interval sequence. An instantaneous heart rate signal was computed from RR interval sequence and the magnitude of the signal was set to the reciprocal of the current interbeat interval. The instantaneous heart rate signal was sampled at 4 Hz and the mean heart rate was subtracted from the resulting one thousand twenty-four point time series. A power spectrum was computed by squaring the absolute value of a Fast Fourier Transform of the one thousand twenty-four point time series. Values for low frequency power (LFP) were computed by integrating the spectrum of between 0.04 and 0.1 Hz. Respiratory frequency power (RFP) was computed by integrating the heart rate power spectrum over a 0.2 Hz-wide band centered at the mean respiratory frequency.
Hard copies of the heart rate time series and power spectrum were printed for each 256 second epochs. Trend graphics for the LFP, the RFP, LFP/RFP ratio, mean heart rate and respiratory rate (hereinafter referred to as the study parameters) were constructed by manually entering data in data files and analyzing the entered data by means of a computer.
Mean values for the study parameters were calculated for each period of study. The Mann-Whitney Rank Sum Test was used to determine statistically significant changes in the study parameters in individual patients and to determine differences among groups of patients. When patients were segregated into more than two groups, the Kruskal-Wallis Test, multiple comparison test, and Tukey's HSD were employed to determine statistical significance. P values of less than, 0.05 were considered significant.
It was found that during each three to twenty-four hour period of study the study parameters for a given patient, the LFP, the RFP and the LFP/RFP ratio (hereinafter referred to as the spectral parameters) remain fairly stable.
Based upon the results of this study, the patients were retrospectively divided into three groups. Group I included seventeen stable patients whose median age was one month. The patients in Group I were without major post-operative complications and did not need prolonged inotropic support. The eight patients in Group II suffered cardiac arrest and died. The median age for the members of Group II was one month. In Group III, there was a total of four patients each of whom was critically ill at the time of the study but later recovered. Median age of the members of Group III was one month. Of the four members of Group III, one required re-operation, one had intermittent hypotensive episodes, and two had cardiac arrests from which they were successfully resuscitated.
In order to separate all twenty-nine patients into a group of stable patients (Group A) and a group of critical patients (Group B), data from each patient in Group III was divided into the data collected during the stable period (which applied to three patients) and the data collected during the preceding critical period (which applied to four patients). When handled in this way, Group A included data for twenty patients and Group B included data for twelve patients. Typical heartrate fluctuation power spectra for Group A and B are respectively illustrated in FIGS. 19 and 20.
In addition, studies were performed on three patients who had isolated coarctation of the aorta at three points in time: upon admission for congestive heart failure; during treatment; during post-operative period; and prior to discharge from an intensive care unit. An attempt was made to identify changes in cardiovascular regulatory function of each of these stages.
Patient profiles for Groups I, II and III are respectively provided in Tables V, VI and VII. These profiles include age, diagnosis and operation.
TABLE V__________________________________________________________________________PATIENT PROFILE; STABLE POST-OP N = 17AGE NO. DIAGNOSIS (NO.) OPERATION__________________________________________________________________________<30 DAYS 9 TGA,IVS (3) ARTERIAL SWITCH TGA,VSD,PS (1) L-BTS HLHS (1) STAGE 1 REPAIR SV (1) L-BTS SEV. COAO (3) SUBCL.FLAP ANGIO.1-12 MO. 5 TGA,IVS (1) ARTERIAL SWITCH TGA,VSD,PS (1) BTS MULT.VSD'S (1) VSD PATCH REPAIR SUPRA-V. PS (1) PA PATCH PLASTY DCRV,VSD,COAO (1) VSD REPAIR, ANOM. B RESECTion1-10 YRS. 2 PS (1) PULM. VALVOTOMY TOF (1) TOF REPAIR>10 YRS. 1 AR,MR AVR,MVR__________________________________________________________________________
TABLE VI__________________________________________________________________________PATIENT PROFILE; CRITICAL, DIED N = 8AGE NO. DIAGNOSIS (NO.) OPERATION__________________________________________________________________________<30 DAYS 4 HLHS (3) NORWOOD PROCEDURE SV W/IAA (1) GORE-TEX GRAFT1-12 MO. 3 HLHS (1) Fontan operation DORV,TAPVC, (1) TAPVC REPAIR, SYS. CCAVC PULM. SHUNT HLHS (1) NON-OPERATIVE61/2 YRS. 1 T OF S/P REPAIR W/ NON-OPERATIVE CHRONIC SEV. CARDIOMYOPATHY, S/P ARREST__________________________________________________________________________
TABLE VII__________________________________________________________________________PATIENT PROFILE: CRITICAL, RECOVERED N = 4AGE NO. DIAGNOSIS (NO.) OPERATION__________________________________________________________________________<30 DAYS 3 HLHS,COAO (1) NORWOOD PROCEDURE HLHS (2) NORWOOD PROCEDURE14 YRS. 1 ACUTE MYOCARDITIS, NON-OPERATIVE S/P ARREST__________________________________________________________________________
In Tables V, VI and VII: TGA is Transposition of the Great Arteries; IVS is Ventricular Septal Defect; PS is Pulmonic Stenosis; HLHS is Hypoplastic Left Heart Syndrome; SV is Single Ventricle; SEV. is severe; COAO is Coarctation of the Aorta; MULT is multiple; VSD is Ventricular Septal Defect; Supra-V. is Supravalulvar; DCRV is Double Chamber Right Ventricle; TOF is Tetralogy of Fallot; AR is Aortic Regurgitation; MR is Mitral Regurgitation; W/IAA is with Interrupted Aortic Arch; DORV is Double Outlet Right Ventricle; TAPVC is Total Anomalous Pulmonary Venous Connections; CCAVC is Complete Common Atrial Ventricular Canal; S/P is Status Post; L is Left; BTS is Blailock Taussig Shunt; PA is Pulmonary Artery; ANOM. is Anomalous; B is muscle Bundle; PULM is Pulmonary; and SYS is Systemic.
Statistically significant differences were observed in the heart rates spectral parameters between the groups of patients as well as among the individual patients. However, the mean heart rate alone did not distinguish stable from critically ill patients. Both the LFP and the LFP/RFP ratio discriminated between the Group A (stable) patients and the Group B (critical) patients. The LFP/RFP ratio grew out of a statistically significant (p less than symbol 0.00001) discrimination between stable and critical patients. Table VIII presents means of study parameters.
TABLE VIII__________________________________________________________________________MEANS OF STUDY PARAMETERS STD. ERROR 99% CONFIDENCE MEAN STD. DEV. OF MEAN LOWER UPPER__________________________________________________________________________GROUP A STABLEPARAMETER(BEATS/MIN.)LFP 1.77 3.35 0.75 -.37 3.91RFP 0.28 0.70 0.16 -.17 0.72LFP/RFP RATIO 8.77 4.86 1.09 8.76 8.79HEART RATE 139 19.60 4.38 139 139GROUP B, CRITICALPARAMETERLFP .05 .03 .01 .02 .07RFP .10 .09 .03 .01 .18LFP/RFP RATIO .83 .51 .15 .83 .83HEART RATE 142 24.32 7.02 142 142__________________________________________________________________________
The discriminate value for the LFP/RFP ratio was two. In Group A, the range of LFP/RFP ratios was 3 to 22 (arithmetic mean 8.77). The range of RFPs was 0.01 to 3.13 (arithmetic mean 0.28) and the range of LFPs was 0.09 to 13.88 (arithmetic mean 1.77). In Group B, the range of LFP/RFP ratios was 0.17 to 1.9 (arithmetic mean 0.83), the ratio of RFPs was 0.02 to 0.32 (arithmetic mean 0.1), and the range of LFPs was 0.01 to 0.1 (arithmetic mean 0.5).
Although the mean value of the LFP/RFP ratio was greater than two for Group I, the ratio for the stable patients fell below two for brief periods. That which distinguishes the stable from the critical patients is the sustained value for greater than or about one hour of the LFP/RFP ratio for the critical group.
The results are graphically depicted in FIGS. 16, 17 and 18. In FIGS. 16 and 17, each heavy dot A represents a geometric mean, each light line B indicates the standard error of the geometric mean and each heavy line C represents the standard deviation of the geometric mean. In FIG. 18, each heavy dot A represents an arithmetic mean, each set of slashes B1 and B2 represents the standard error of the arithmetic mean and each set of slashes C1 and C2 represents the standard deviation of the arithmetic mean.
The significance of heart rate spectral analysis for diagnosis of cardiovascular stress and the prediction of fatality is highlighted by the fact that patients with a low LFP/RFP ratio underwent a cardiac arrest even in the presence of otherwise normal vital signs. No patient with a LFP/RFP ratio greater than two experienced a cardiac arrest.
Infusion of pressors, alone or in combination with vasodilators, did not induce a low LFP/RFP ratio.
Four patients in Group III had LFP/RFP ratios less than two during their critical periods. For the three of these four patients who were restudied during their recovery periods, all three had LFP/RFP ratios greater than two.
The mean LFP for Group B [0.05 (Beats per minute).sup.2 ] was less than the mean LFP for Group A [1.77 beats per minute).sup.2 ], p<0.0001. There was no significant difference between the mean RFP between the groups.
The initial LFP/RFP ratios for the patients with isolated coarctation of the aorta ranged up to 10 000. The LFP/RFP ratios observed for this group immediate after an operation to correct the condition were within the range for Group A patients. Two patients had LFP/RFP ratios greater than 100 before discharge from the intensive care unit. These ratios were correlated with mild to moderate congestive heart failure. One of these patients died suddenly at approximately 21/2 months after the operation. The other two patients remained alive and well.
Although the LFP/RFP ratio provided the sharpest discrimination between stable and critical patients in these studies, the LFP alone discriminated between Groups A and B, p<0.0001. Neither respiratory frequency peak power nor mean heart rate distinguished between Groups A and B. On the other hand, LFP/RFP ratios and LFP low levels sustained for greater than or about one hour correlate with the course of the conditions of patients who experienced cardiac arrest or severe hypotensive episodes but later recovered.
Although stable patients experienced transient depression of levels of LFP and of the LFP/RFP ratio, depression of these factors for about an hour or more never failed to predict a critical status.
No significant difference was observed between freely ventilating patients and mechanically ventilated patients. Eighteen out the twenty patients in Group A were mechanically ventilated and all twelve of the Group B patients were mechanically ventilated.
All patients in Group B received inotropic support while more than half of the patients in the Group A received at least some inotropic support. The cardiac diagnoses of all of the patients in Group B and for some of the patients in Group A were known to be associated with high mortality. All of the patients in Group B underwent deep hypothermic circulatory arrest during their operations. Of the twenty patients in Group A, nine had extra cardiac surgery (i.e. not involving cardiopulmonary bypass or deep hypothermic circulatory arrest). Three of the patients in Group II did not undergo operations. Therefore, it is not believed that differences in treatment or disease specific pathology alone explained the low values LFP and the low LFP/RFP ratios in Group B patients but that the low values actually reflect a vulnerable circulatory state.
It has also been observed that the value of LFP and of the LFP/RFP ratio increase in moderate to severe heart failure but decreased to subnormal values in end stage myocardial failure. Thus, these two spectral parameters may indicate cardiovascular regulatory effectiveness (cardiovascular regulatory reserve) during the stress of heart failure.
This analysis is consistent with previous physiological studies which indicated that low frequency heart fluctuations may be mediated by both the beta-sympathetic and parasympathetic mechanisms while respiratory fluctuations are exclusively mediated by parasympathetic mechanisms. It is also consistent with this analysis that LFP has been observed to increase during conditions which elicit enhanced sympathetic activity, such as acute hypoxia, postural changes, hemorrhage and aortic constriction. In this light, the LFP/RFP ratio may represent a measure of the balance between beta adrenergic and parasympathetic modulation of cardiac function.
Thus, the increase in LFP and in the LFP/RFP ratio for patients with isolated coarctation of the aorta and moderate heart failure may result from an increased activity from the sympathetic mechanism and a decreased activity of the parasympathetic mechanism. On the other hand, the decreased level of LFP and of the LFP/RFP ratio found in critical patients may be due to non-responsiveness of the sympathetic mechanism. Sympathetic non-responsiveness may be due to myocardial catecholamaine depletion alone or in combination with the observed down regulation of beta receptors from cardiac tissue in the end stage of heart failure.
EXAMPLE II
In patients undergoing operations, shifts in body fluid disposition during surgery may lead to changes in intervascular volume (i.e. a shift of fluid out of a circulatory tree of blood vessels). Accordingly, the availability of the method of diagnosing cardiovascular stress as described in Example I may be used to choose among various protocols for treatment or to justify a radical change in medical or surgical treatment.
For example, by monitoring a patient with the real time heart rate frequency spectral monitor according to the present invention during administration of anesthesia, an anesthesiologist may non-invasively monitor intravascular volume status. Upon observing an increase in the LFP or in the LFP/RFP ratio, the anesthesiologist may increase the amount of fluids administered by way of intravenous injection or may take steps to reverse effects of a particular anesthetic.
It is a particular advantage of the apparatus according to the present invention that heart rate fluctuation spectral analysis may be done in real time. This capability permits correlation of treatment administered with changes in LFP or LFP/RFP ratios.
Although the present invention has been described in terms of preferred embodiments, it is understood that modifications, variations and improvements will occur to those skilled in the art. For example, it will occur to those skilled in the art to employ the present invention for monitoring cardiovascular instability in the following settings in which significant circulatory stress are commonly observed: Labor and Delivery Room; Operating Room; Cardiac Catheterization Laboratory; Neonatal, Pediatric, Adult Medical, Adult Surgical, Cardiothoracic and Neurosurgical Intensive Care Units; Coronary Care Units; Burn Units; and Emergency Rooms.
The present invention may also be used for monitoring cardiovascular instability in the following patients in which adjustments in cardiovascular regulation may provide a central key to understanding the efficiency and efficacy of treatment. Ambulatory patients with known heart disease in which sudden cardiac death is a common association, one example of which would be a patient with a congestive cardiomyopathy who is being treated with vasodialator drugs and for whom the LFP/RFP ratio has changed from a normal baseline level to decreased levels may then subsequently be either admitted to the hospital for adjustment of medications and/or observed and monitored in the physican's office while his vasodialator drug dose is increased. A patient with renal disease (e.g. one who requires dialysis) may exhibit a marked increase in LFP and LFP/RFP ratio secondary to the onset of incipient moderate congestive heart failure would thus be treated by dialysis to relieve a congested circulatory state; a patient with moderate to severe pulmonary disease resulting in hypoxemia and/or hypercarbia who requires bronchodialator and/or supplementary oxygen and/or mechanical ventilation (e.g. a patient who exhibits a marked decrease in LFP/RFP ratio secondary to myocardial failure due to a profound imbalance between myocardial ventricular output and oxygen demand), may be treated by adjustments in bronchodialator drugs, diuretics, and/or ventilator adjustments.
A premature infant of very low birth weight known to be at risk for intraventricular hemorrhage may, for example, develop a slow intracranial bleed associated with an abrupt increase in LFP, which may alert physicians prior to a brisk bleed thus allowing institution of appropriate changes in medical management to limit substantially known risk factors that may predispose to such an event, or may permit recognition of the presence of unsuspected circumstances that contribute to the bleed. In neurologic disease, such as one in which a patient has sustained a major intracerebral event (e.g. neurosurgical evacuation of a space occupying lesion such as a tumor or blood), a patient may, for example, exhibit a markedly attenuated LFP/RFP ratio, secondary to massively increased parasympathetic activity which would markedly increase RFP, at the expense of LFP, but which may or may not be associated with signs of increased intracranial pressure, and which may be treated by, for example, hyperventillation, rapid diuresis, or burr hole placement.
A patient with severe systemic infection may exhibit shock secondary to the infection process may, for example, exhibit an elevated LFP/RFP ratio which may then be subsequently used by the physician in managing the shock state by means of pressor agents and infusion of significant volumes of fluid, thus providing the physician an indication of how effectively he is treating the shocked state above and beyond the traditional measurements such as systemic blood pressure and cardiac output. A patient with hematologic disease associated with anemia, such as Sickle Cell Anemia, exhibits an oscillation in capillary blood flow when severely anemic at the frequency associated with LFP and may exhibit large values for LFP, and for the LFP/RFP ratio may, for example, be treated by blood transfusion which may lead to an expected decrease in LFP, LFP/RFP ratio, and thus enable the physician to monitor by means of heart rate spectral analysis appropriate timing for transfusion therapy. A fetus prior to delivery, may for example, exhibit a marked attenuation in LFP associated with severe fetal distress, and may thus alert the physician to perform an emergency Caesarean section.
One skilled in the art understands that the calibrators according to the present invention may be adjusted to simulate disease states as well as normal conditions. It is also understood that the present invention is not limited to use with patients whose primary disease is of the heart but that modifications may be made for use with such patients.
Lastly, it is clear to one skilled in the art that durations and ranges for levels of LFP and LFP/RFP ratios are conservatively stated herein and that variations from these ranges and durations are contemplated within the scope of the equivalents of the present invention.
Therefore, it is intended that the methods and apparatus according to the present invention to be given the broadest scope allowable for the invention as claimed.
The following appendices are program listings for software useful in implementing the present invention. Specifically: Appendix A illustrates software for an embodiment of the present invention applicable to a Hewlett-Packard microcomputer; Appendix B illustrates software according to the present invention useful for practicing the present invention on an IBM personal computer; and Appendix C is a calibration program designed to implement the software driven calibrator of FIG. 11. ##SPC1##

Number	Name	Date
3612041	Ragsdale	Oct 1971
3698386	Fried	Oct 1972
4170992	Dillman	Oct 1979
4289141	Coanier	Sep 1986
4379460	Judell	Apr 1983
4422458	Kravath	Dec 1983
4432375	Angel et al.	Feb 1984
4463764	Anderson et al.	Aug 1984
4506678	Russell et al.	Mar 1985
4519395	Hrushesky	May 1985
4545387	Balique	Oct 1985
4549551	Dyck et al.	Oct 1985
4570225	Lundy	Feb 1986
4573478	Arnold	Mar 1986

Methods and apparatus for monitoring cardiovascular regulation using heart rate power spectral analysis

Information

Patent Number

Date Filed

Date Issued

Inventors

Original Assignees

Examiners

Agents

CPC

US Classifications

Field of Search

US

International Classifications

Abstract

Description

Claims

US Referenced Citations (14)

Foreign Referenced Citations (1)

Non-Patent Literature Citations (23)

Entry
"Computer Electrocardiographic Processing", Cesar A. Caceres, Aug. 1972.
"Non Invasive Monitoring Techinques in Neurosurgical Intensive Care", James J. Ackmenn.
Baselli et al., Med. Inform., 10, 223-235 (1985).
Bhargava et al., J. Electrocardiol., 14, 57-60 (1981).
Cary, "A Digital Cardiac Tachometer Synthesizer and Signal Delay", B.S Thesis, Massachusetts Institute of Technology (1983).
de Boer et al., Med. Biol. Eng. Comput., 23, 352-358 (1983).
de Leon et al., Abstract presented at a Reynolds Symposium, Jun. 5, 1984.
Fischler, "Real-Time Microprocessor-Based Analysis of Heartbeat Variability", B.S. Thesis, Massachusetts Institute of Technology (1982).
Fiser et al., Automedica, 2, 143--147 (1978.
Frank, "Analysis of Heart-Rate Variability Using the Radio Shack Color Computer", B.S. Thesis, Massachusetts Institute of Technology (1983).
Gordon et al., Pediatr. Research, 18(10) 921-926 (1984).
Jennings et al., Exp. Aging Research, 10, 19-23 (1984).
Kuo et al., Computers in Cardiology, 347-349, Sep. (1978).
Nygards et al., Computers in Cardiology, 393-398, Sep. (1977).
Pagani et al., Eur. J. Clin. Invest., 14, 19 (1984).
Penaz et al., Physiologia Bohemoslovaca, 27, 349-357 (1978).
Putman et al., Journal of Clinical Engineering, 5(1), 56-58, Jan.-Mar. (1980).
Raghuveer, "Bispectrum and Multidimensional Power Spectrum Estimation Algorithms based on Parametric Models with Applications to the Analysis of ECG Data," Ph.D. Dissertation, University of Connecticut (1984).
Schick et al., Ann. Biomed. Eng., 6, 154-160 (1978).
Tu, "Time Spectral Analysis of Physiological Signals," Microprocessor System for Real, M.S. Thesis, Massachusetts Institute of Technology (1984).
Wilson et al., Med. & Biol. Eng. & Comput., 20, 293-298 (1982).
Zwiener, Automedica, 2, 161-169 (1978).
Zwiener et al., Automdeica, 5, 77-90 (1984).