Estimation of blood flow and hemodynamic parameters from a single chest-worn sensor, and other circuits, devices and processes   

This application is related to U.S. patent applications as follows:

This application is related to U.S. patent application: “Heart Monitors And Processes With Accelerometer Motion Artifact Cancellation, And Other Electronic Systems” Ser. No. 12/______ (TI-68518) filed Aug. 24, 2010 simultaneously herewith, for which priority is claimed under 35 U.S.C. 120 and all other applicable law, and which is incorporated herein by reference in its entirety.

This application is related to U.S. patent application “Motion/Activity, Heart-Rate And Respiration From A Single Chest-Worn Sensor, Circuits, Devices, Processes And Systems” Ser. No. 12/______ (TI-68552) filed Aug. 24, 2010 simultaneously herewith, for which priority is claimed under 35 U.S.C. 120 and all other applicable law, and which is incorporated herein by reference in its entirety.

This application is related to provisional U.S. patent application “Motion Artifact Cancellation to Obtain Heart Sounds from a Single Chest-Worn Accelerometer” Ser. No. 61/242,688 (TI-68518PS) filed Sep. 15, 2009, for which priority is claimed under 35 U.S.C. 119(e) and all other applicable law, and which is incorporated herein by reference in its entirety.

This application is related to provisional U.S. patent application “Motion/Activity, Heart-rate and Respiration From a Single Chest-worn Sensor” Ser. No. 61/262,336 (TI-68552PS) filed Nov. 18, 2009, for which priority is claimed under 35 U.S.C. 119(e) and all other applicable law, and which is incorporated herein by reference in its entirety.

This application is related to provisional U.S. patent application “Estimation of Blood Flow and Hemodynamic Parameters from a Single Chest-worn Sensor” Ser. No. 61/262,331 (TI-68553PS) filed Nov. 18, 2009, for which priority is claimed under 35 U.S.C. 119(e) and all other applicable law, and which is incorporated herein by reference in its entirety.

This application is related to provisional U.S. patent application “Heart Rate Detection In High Noise Conditions” Ser. No. 61/104,030 (TI-66732PS) filed Oct. 9, 2008, for which priority is claimed under 35 U.S.C. 119(e) and all other applicable law, and which is incorporated herein by reference in its entirety.

This application is related to U.S. patent application Publication “Heart Rate Detection In High Noise Conditions” 20100094150, dated Apr. 15, 2010 (TI-66732) for which priority is claimed under 35 U.S.C. 120 and all other applicable law, and which is incorporated herein by reference in its entirety.

This application is related to provisional U.S. patent application “Robust Heart Rate Detection in the Presence of Pathological Conditions” Ser. No. 61/023,581, filed on Jan. 25, 2008 (TI-65798PS), for which priority is claimed under 35 U.S.C. 119(e) and all other applicable law, and which is incorporated herein by reference in its entirety.

This application is related to U.S. patent application Publication “Method and System for Heart Sound Identification” 20090192401, dated Jul. 30, 2009 (TI-65798) for which priority is claimed under 35 U.S.C. 120 and all other applicable law, and which is incorporated herein by reference in its entirety.

This application is related to U.S. patent application Publication “Method and Apparatus for Heart Rate Monitoring” Ser. No. 12/768,488 filed Apr. 27, 2010 (TI-67877), which is incorporated herein by reference in its entirety.

This application is related to U.S. patent application “Parameter Estimation for Accelerometers, Processes, Circuits, Devices and Systems” Ser. No. 12/398,775 (TI-65353) filed Mar. 5, 2009, and which is incorporated herein by reference in its entirety.

This application is related to the US patent application titled “Processes for More Accurately Calibrating E-Compass for Tilt Error, Circuits, and Systems” Ser. No. 12/398,696 (TI-65997) filed Mar. 5, 2009, and which is incorporated herein by reference in its entirety.

Not applicable.

Portions of this patent application contain materials that are subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document, or the patent disclosure, as it appears in the United States Patent and Trademark Office, but otherwise reserves all copyright rights whatsoever.

The field of technology is in the areas of monitoring of the human body, automatic analysis and display of monitoring data locally for medical and other purposes and telecommunication remotely for tele-medicine, and processes, circuits and devices for body monitoring of heart function, circulatory function, respiration, or other physiological processes. Biomedical instrumentation and signal processing are further fields.

Ambulatory measurement of cardiac activity can facilitate home health monitoring of older adults and of patients with a history of cardiovascular conditions. Evaluating cardiovascular performance of patients in ICU (intensive care unit) and hospital settings, in mobile ambulances, and at accident and trauma sites also involves or can involve ambulatory cardiac measurement.

Most current solutions for heart rate monitoring involve cumbersome equipment, such as heart rate recording belts to be worn around the chest, electrocardiogram (ECG) electrodes and leads, and in most cases electrical contact to the skin. However, such methods remain obtrusive, and are not optimal for long-term and ambulatory monitoring.

An alternative method of heart rate measurement uses heart sounds, conventionally measured with stethoscopes or phonocardiograph.

Detection and early warning of risk factors for and any incident of heart failure is vitally important in medicine, allied medical fields, residential care-giving, exercise venues and other settings. Heart failure can be caused by, and is at risk in case of, coronary artery disease, hypertension, valve disorder, past myocardial infarction, muscle disorder, congenital heart conditions, etc.

Current solutions for not only heart rate monitoring but also respiration monitoring are believed to involve cumbersome and expensive equipment e.g., respiration and heart rate monitoring belts to be worn around the chest, spirometers and canulas to be worn around the mouth and nose, and electrocardiogram (ECG) electrodes and leads to be taped on the body. Not only are these solutions obtrusive and expensive, but may also be too restrictive to be well-suited for ambulatory monitoring.

Noise mixed with signals received by the sensors used in heart monitoring, respiration monitoring, body motion and other monitoring applications can adversely affect the accuracy of each type of signal. Accordingly, methods for robust detection and separation of such signals in noisy conditions are desirable. Accuracy of heart rate detection is important in many commercial heart monitoring applications (e.g., heart rate monitors in exercise equipment, personal heart rate monitors, etc.) and medical heart monitoring applications (e.g., digital stethoscopes, mobile cardiac monitoring devices, etc.).

Simpler, more economical and more efficient methods and devices are desirable in the art for obtaining, isolating, determining and monitoring resting data and ambulatory data, such as robust, accurate detection of heart rate, timings of heart sounds (S1 and S2) and pathological cardiac conditions, and robust detection of respiration in connection with respiratory and pulmonary disorders, as well as data on body motion and ambulatory data and activity data.

Conventional approaches to address the bodily motion signal separation and/or removal problem are believed to involve multi-signal adaptive algorithms that need an additional motion signal reference recording typically from a secondary sensor. Also, the reference signal needs to be reasonably well correlated to the motion picked up by the primary sensor. Such arrangements are very difficult to establish in a real setting and can cause poor rejection of the motion signal and body motion artifacts. Some conventional single-channel de-noising techniques reinforce all major signal peaks and fail to distinguish body motions from heart sounds.

In addition to medical-related applications, solving the above problems could also help monitor older adults for unexpected changes in gait, for falls, for syncope (fainting), for accidents and trauma incidents. Fitness monitoring at home, in exercise venues, and in institutional care settings could also benefit.

Hemodynamic data also challenge the art to find methods and devices for obtaining, isolating, determining and monitoring more simply, economically and more efficiently. Hemodynamics as discussed herein includes the study of blood flow-related data directly or indirectly related to blood flow, such as: heart stroke volume, cardiac output, pre-ejection period, contractility (ability of heart to contract, inotropy), and related causal or caused bodily dynamics such as exercise and exercise recovery, and the Valsalva maneuver (such as when pushing or straining while holding one\'s breath, or otherwise doing the maneuver in a medical test).

Measurement of blood flow, hemodynamics and cardiovascular performance is integral to a holistic assessment of an individual\'s health. Specifically, patients with past conditions of heart disease like heart failure (potentially arising out of one or more of many causes like coronary artery disease, heart valve or heart muscle disorders, past myocardial infarction, hypertension etc.) may need constant monitoring in order to improve a person\'s quality of life via timely and appropriate diagnostic interventions. While the physiological mechanisms underlying these conditions are fairly well understood, the technology to monitor these physiological vitals needs considerable improvement.

Most current solutions for the measurement of blood flow and other hemodynamic parameters are believed to involve cumbersome and expensive equipment e.g., Impedance Cardiography (calls for electrodes to be connected on the skin), Doppler Echo Cardiography, Continuous Blood Pressure Monitoring etc. Not only are these solutions obtrusive and expensive, but may also be too restrictive to be well-suited for ambulatory monitoring applications.

SUMMARY

Generally, and in one form of the invention, an electronic monitoring device includes an electronic processor having at least one signal input for body monitoring, and a memory holding instructions for the electronic processor coupled to the electronic processor so that the electronic processor is operable to isolate a cardiac signal including cardiac pulses combined with other cardiac signal variations, and the electronic processor further operable to execute a filter that separates a varying blood flow signal from the cardiac pulses and to output information based on at least the varying blood flow signal.

Generally, and in another form of the invention, an accelerometer sensor assembly has a broadside portion and includes an accelerometer sensor circuit having an axis of acceleration sensitivity parallel to the broad side and orientable on the chest to deliver an input signal representing a component of acceleration approximately parallel to a head-to-feet body axis and including heart pulses and other variations mixed together, an electronic circuit responsive to the accelerometer sensor circuit and operable to execute a filter that delivers a varying blood flow signal from the input signal at least when that axis of acceleration sensitivity is approximately parallel to the head-to-feet body axis, the varying blood flow signal substantially freed of heart pulses, and an output circuit operable to send information based on the varying blood flow signal.

Generally, and in a process form of the invention, an electronic monitoring process includes electronically bandpassing digital signals representing living-body monitoring signals in a range selective for a cardiac signal, and executing a filter that delivers a varying blood flow signal from the cardiac signal.

Generally, and in a further form of the invention, a hemodynamic monitoring device includes an electronic processor having at least one signal input for body monitoring, and a memory holding instructions for the electronic processor coupled to the electronic processor so that the electronic processor is operable to obtain a cardiac pulse signal having a varying amplitude, and the electronic processor further operable to post-process the cardiac pulse signal to provide a time-varying output representing an estimation for at least one hemodynamic parameter based on the amplitude and selected from the group consisting of stroke volume SV and cardiac output CO.

Other devices, sensor assemblies, electronic circuit units, and processes are also disclosed and claimed.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a partially-block, partially-pictorial, partially graphical depiction of an inventive structure and process for separating a heart signal from body motion and noise using a single chest sensor.

FIG. 2 is a partially-schematic, partially-pictorial, partially graphical depiction of a measurement setup including both an accelerometer sensor on the chest with a composite signal having heart signals and motion and noise, as well as an electrocardiogram ECG circuit with ECG electrodes affixed to the body and showing an operational amplifier and an ECG signal.

FIG. 3 is a block diagram of an inventive structure and process for separating a heart signal from body motion and noise using a single accelerometer chest sensor.

FIG. 4 is a flow diagram of the inventive structure and process for separating a heart signal from body motion and noise using a single accelerometer chest sensor and remarkable smoothing filter and residue circuit, envelope-based noise rejection, folded correlation and other steps.

FIG. 5 is a set of four concurrent waveform traces of voltage versus time in various parts of the inventive structure and process of FIGS. 2, 3 and 4 with a subject walking around a room. A time interval portion of the traces is magnified and shown as four time-magnified waveforms maintaining the same voltage scale for each. Some of the traces are accelerometer-based and one is ECG-based.

FIG. 6 is a pair of concurrent accelerometer-based waveform traces of voltage versus time in parts of the inventive structure and process of FIGS. 2, 3 and 4 with a subject walking on a treadmill. A time interval portion of the traces is magnified and shown as time-magnified waveforms maintaining about the same voltage scale for each.

FIGS. 7A, 7B, and 7C are each a pair of concurrent noisy handgrip ECG-based waveform traces of voltage versus time in parts of the inventive structure and process of FIGS. 2 and 4 with a subject walking on a treadmill. The ECG-based waveforms are one unfiltered, and one inventively filtered to recover a heart sounds signal.

FIG. 8 is a block diagram of another inventive structure and process for obtaining cardiac information using an accelerometer chest sensor, such as heart rate and information related to beat-to-beat changes in stroke volume and cardiac output.

FIG. 9 is a pair of concurrent accelerometer-based waveform traces of voltage versus time in parts of the inventive structure and process of FIGS. 3 and 4 pertaining to envelope-based noise rejection.

FIG. 10 is a pair of concurrent accelerometer-based waveform traces of voltage versus time in parts of the inventive structure and process of FIGS. 3 and 4 pertaining to folded correlation.

FIG. 11 is a plot of the difference of two heart-rate measures (inventively filtered accelerometer-based and ECG-based) versus their average.

FIG. 12 is a plot having time interval between adjacent S1 cardiac pulses from inventively filtered accelerometer data on one graph axis, versus ECG R-R interval on the other graph axis.

FIG. 13 is a partially-block, partially-pictorial, partially graphical depiction of an inventive structure and process for separating a respiration signal from heart and body motion and other signals using a single chest sensor.

FIG. 14 is a partially-flow, partially graphical depiction of an inventive process for FIG. 13 separating a respiration signal, a heart signal and a body motion signal from each other using a single chest sensor.

FIG. 15 is a pair of concurrent accelerometer-based waveform traces of voltage versus time in parts of the inventive structure and process of FIGS. 13-14 and shows a raw and inventively filtered accelerometer-based signal during rest and brisk motion. A time portion of the signals during rest is magnified in both voltage scale and time scale. A time portion of the signals during subsequent motion is magnified in time scale and not voltage scale.

FIG. 16 is a set of four concurrent waveform traces of voltage versus time in various parts of the inventive structure and process of FIGS. 13-14 during motion and brisk walking A time interval portion of the traces is magnified and shown as four time-magnified waveforms maintaining the same voltage scale for each. Some of the traces are inventively filtered accelerometer-based and one is ECG-based.

FIG. 17 is a flow diagram of a process for FIGS. 13-14 to separate a respiration signal from a heart signal and using inter-beat intervals of the heart signal, with both the respiration signal and the heart signal substantially separated from body motion and noise signals.

FIG. 18 is a set of three concurrent waveform traces of voltage versus time in various parts of the inventive structure and process of FIGS. 13-14, showing raw signal and inventively-obtained residue from filtered accelerometer-based signal, and further showing a respiration signal generated from ECG—R-R interval, and accelerometer heart sounds S1-S1 interval of the residue signal. A time interval portion of the traces is magnified and shown as three time-magnified waveforms maintaining the same voltage scale for the first two, and magnifying the voltage scale for the respiration signal.

FIG. 19 is a flow diagram of another process for FIGS. 13-14 to separate a respiration signal from a heart signal according to baseline wander method herein for respiration monitoring by a single inventively filtered accelerometer sensor.

FIG. 20 is a set of four example waveforms of voltage versus time and shown for comparison of a respiration belt signal with respiration outputs from each of the processes of FIGS. 17, 19 and 22.

FIG. 21 is a set of seven example waveforms of voltage versus time, including a comparison of a reference respiration signal and ECG-derived respiration signal with respiration outputs from each of the processes of FIGS. 17, 19 and 22.

FIG. 22 is a flow diagram of a process for FIG. 13 to inventively separate a respiration signal from a heart signal by amplitude modulation detection of peak heights of the heart signal.

FIG. 23 is a set of three concurrent waveform traces of voltage versus time in various parts of the structure and process of FIGS. 2 and 13 and 22, showing ECG amplitude modulation on the R peaks, and further showing amplitude modulation on the S1 peaks from inventively filtered accelerometer based sensing, and also showing a respiration signal obtained from a respiration belt for reference.

FIG. 24 is a block diagram of an inventive wired system structure and process and including inventive structures and processes from the other Figures.

FIG. 25 is a block diagram of an inventive wireless system structure and process and including inventive structures and processes from the other Figures.

FIG. 26 is a partially-block, partially-pictorial, partially graphical depiction of an inventive structure and process for separating a blood flow signal from heart and other signals using sensor signals from one or more axes of a single chest sensor.

FIG. 27 is a pair of concurrent accelerometer-based waveform traces of voltage versus time of sensor signals from multiples axes of a single chest sensor in the inventive structures and processes of FIG. 26.

FIG. 28 is a voltage-versus-time graph of a pair of concurrent accelerometer-based waveforms from Z- and Y-axes of the single chest sensor, along with inventively filtered Y-axis signal and residue in the inventive structures and processes of FIG. 26.

FIG. 29 is a voltage-versus-time graph of three concurrent waveforms including a pair of inventively filtered accelerometer-based waveforms from Z- and Y-axes of the single chest sensor in the inventive structures and processes of FIGS. 26, 30 and 31, compared with a reference ECG waveform.

FIG. 30 is a partially-block, partially-pictorial, partially graphical depiction of another inventive structure and process for separating a blood flow signal and hemodynamic parameters, respiration signals, heart signals and motion signals from each other using sensor signals from one or more axes of a single chest accelerometer sensor.

FIG. 31 is a combined flow diagram of inventive processes for separating a heart signal from body motion and noise using Z-axis sensor input and separating a blood flow signal using Y-axis sensor input from FIG. 30 and further electronically processing the heart signal and blood flow signal jointly to generate hemodynamic parameter signals for a display as in FIGS. 24 and 25.

FIG. 32 is a voltage-versus-time graph of three concurrent waveforms including an ECG signal, a filtered heart signal from the Z-axis accelerometer sensor, and a blood flow signal filtered from Y-axis of the accelerometer sensor in the inventive structures and processes of FIGS. 26, 30 and 31, and further showing time locations P1 and F1 and hemodynamic parameters for isovolumic contraction interval IVCI, pre-ejection period PEP and flow peak amplitudes PAmp and Jamp.

FIG. 33 is a graph of voltage (arbitrary units) versus time-samples for a multitude (ensemble) of waveforms each of a respective instance of inventively filtered blood flow signal from Y-axis of the accelerometer sensor in the inventive structures and processes of FIGS. 26 and 30. (FIG. 33 is on same sheet as FIG. 37.)

FIG. 34 is a voltage-versus-time graph of four concurrent waveforms during exercise recovery, the waveforms including reference ECG, inventively filtered blood flow signal from Y-axis, PEP, and PAmp. A time interval portion of the traces is magnified and shown as four time-magnified waveforms maintaining the same voltage scale for each except for modest voltage scale magnification for PEP and PAmp.

FIG. 35A is a voltage-versus-time graph of four concurrent waveforms over about a minute for a Valsalva Release phase of a Valsalva maneuver; the first waveform representing inventively-produced residue from polynomial filtering the accelerometer Z-axis as in FIG. 31 (left side), the second waveform representing peak amplitude PAmp of that residue, the third waveform representing Stroke Volume, and the fourth waveform representing Cardiac Output.

FIG. 35B is a voltage-versus-time graph of another four concurrent waveforms over about a minute for a Valsalva Release phase of a Valsalva maneuver; the first waveform representing the blood flow signal from inventive polynomial filtering of the accelerometer Y-axis as in FIG. 31 (right side), the second waveform representing peak amplitude PAmp of that blood flow signal from accelerometer Y-axis, the third waveform representing Stroke Volume, and the fourth waveform representing Cardiac Output.

FIG. 36A is a flow diagram of inventive process for separating a heart signal from body motion and noise using Z-axis sensor input such as for use in FIG. 31 or with FIG. 36B.

FIG. 36B is a flow diagram of inventive process for separating a heart signal as well as a blood signal from each other using Y-axis sensor input such as for use in FIG. 31 and for obtaining further hemodynamic data and other information from the single Y-axis of the chest sensor.

FIG. 37 is a model of a standing subject, the model described by a second-order differential equation to approximate the blood flow signal of the standing subject as a solution thereof.

FIG. 38 is a model of a subject lying prone, the model described by a second-order differential equation having different model parameters than in FIG. 37, to approximate the blood flow signal of the prone subject as a solution thereof.

FIG. 39 is a block diagram of a system structure for use in and improved according to inventive structures and processes from the other Figures.

FIGS. 40A and 40B are respective broadside and cross-sectional views of an inventive accelerometer sensor and transponder chip mounted on a support plate affixed by an adhesive tape to the chest, and for use with the inventive structures and processes from the other Figures.

FIG. 41 is a block diagram of an inventive structure and process for variably combining accelerometer signals from multiple axes in various proportions to provide one or more inputs to the smoothing filtering of FIGS. 31, 36A and 36B and various other Figures herein.

FIG. 42 is a voltage-versus-time graph of four concurrent waveforms including reference ECG, acceleration along a dorso-ventral axis (Z-axis), acceleration along a superior-inferior axis (Y-axis) and acceleration along a dextro-sinistral axis (X-axis), the various acceleration signals for use in the circuit FIG. 41 and circuits of other Figures.

Corresponding numerals in different Figures indicate corresponding parts except where the context indicates otherwise. A minor variation in capitalization or punctuation for the same thing does not necessarily indicate a different thing. A suffix .i or .j refers to any of several numerically suffixed elements having the same prefix. A first, second, third, etc. waveform is referenced in top to bottom order for a given Figure.

DETAILED DESCRIPTION

Some structure and process embodiments provide motion artifact cancellation or motion signal separation to obtain heart sounds from a single chest-worn accelerometer.

Miniature, high-sensitivity MEMS accelerometers are presently available. Here, such an accelerometer is incorporated into a single, chest-worn sensor for recording of signals including some related to heart sounds. (The latter signal components are also themselves sometimes called heart sounds herein. The term “heart sound” refers in an expansive way to a signal analogous to cardiac S1, S2, and/or heart murmur or other cardiac waveform features, obtained from the processing of accelerometer data or other sensor data, and not necessarily to an audible sound.)

However, a major challenge of ambulatory monitoring is the corruption of heart signals by body motion artifact signals and the confusion of such signals. In some measurements, the chest acceleration signal as picked up by the accelerometer 10 in FIGS. 1-3 had a rather slow varying, but very strong (20-50 my peak-to-peak) motion component. Riding on top of this motion signal, was a higher frequency, but weaker (5-10 my peak-to-peak) heart sound signal. Significant variability between subjects was observed in the frequency content of both the motion and the heart sounds. Also, the two signals—motion and heart sounds—are not entirely frequency separable. Thus, simple digital band pass filtering does not consistently work to separate them. Physical motion impulses from the feet couple very differently and in a non-stationary and non-correlated manner to sensors placed at different parts of the body and also to orthogonal axes of the same sensor. Accordingly, even using multiple sensors to cancel out an artifact is complicated or unreliable.

Some of the embodiments remarkably introduce a Data Acquisition/Signal Processing unit 20 with a special smoothing filter 130 in FIGS. 3-4 that tracks slow varying body motion signal wander or variation and then removes the wander from the sensor-based signal to give a clean (motion-removed) biomedical signal of interest on a Display unit 30. The smoothing filter 130 involves a polynomial filter or comparably effective smoothing filter used directly or in a composite signal processing path. Some embodiments use a subtraction step 140 as in FIG. 4 to remove non-stationary motion artifacts reliably and robustly. Removing such artifacts makes the system more fully immune to sensor placement and contact variations on the chest that might arise when using sensor 10. This provides a simple, yet effective way to reduce the impact of motion artifacts and allow the reliable detection of primary heart sounds and subsequent derivation of heart rate even when a person is walking while being monitored. In this way, motion signal removal or separation, and heart-sound signal detection and heart-rate detection are facilitated. No secondary reference or noise source is needed, thus reducing complexity of system design. Embodiments of structure and method thus extract primary heart sound signals from chest-worn sensor (e.g., accelerometer) data in the presence of motion artifacts.

Results from six subjects showed a primary heart signal detection rate of 99.36% with a false positive rate of 1.3% as described elsewhere herein (TABLE 2). Such type of embodiment appears to outperform noise removal techniques such as wavelet de-noising and adaptive filtering. (In certain motion conditions, or in combination, alternative approaches like Wavelet Decomposition, Adaptive Filtering, Blind Source Separation may in some embodiments also be used instead of, separately from, parallel to, or in combination with, the polynomial filtering.)

Advantages include: 1) uses as few as a single sensor or signal capture component, 2) eliminates use of a secondary reference sensor, 3) allows unobtrusive and non-invasive monitoring of vital biomedical signals in ambulatory settings for continuous monitoring applications, 4) separates heart signals independent of non-stationary bodily motion wander.

For biomedical instrumentation and signal processing for heart sounds specifically, problematic motion artifacts are thus removed from biomedical signals—such as from chest accelerometer signals and/or from electrocardiogram (ECG) signals—for use in ambulatory health monitoring settings. The embodiments can also be extended by use of a spectrum analyzer (Fourier analysis) to extract frequency separable components of interest too.

Ambulatory monitoring of cardiac activity can find widespread applications in home health monitoring of patients with a history of cardiovascular conditions, monitoring older adults, ICU and hospital monitoring, monitoring vital signs in mobile ambulances, at accident and trauma sites and can be used for fitness monitoring at exercise centers and elsewhere.

In some structure and process embodiments for removal of motion-related artifacts from biomedical signals, beneficial monitoring is provided for, e.g., either or both of two independent signal sources—accelerometer 10 and ECG of FIG. 2. Chest acceleration signals are collected in an ambulatory (walking) setting from real human subjects using a chest-worn accelerometer 10—providing primary heart sounds signified as S1 and S2. Heart sound S1 includes audible sounds concurrent with tricuspid and mitral valve activity and shows on a seismocardiogram as a pulse bundle. Heart sound S2 is mainly associated with pulmonary valve and aortic valve activity. Structures and processes of the embodiments thus remove motion artifacts and facilitate the use of a single, miniature, chest-worn MEMS accelerometer to pick up heart activity and heart rate—derived from heart sounds—from ambulatory subjects as shown in FIGS. 1 and 2. In FIG. 2, electrocardiogram signals are independently collected from the human subjects walking briskly or running on a treadmill—providing signal components such as QRS from the ECG.

Some background heart anatomy terms are as follows. De-oxygenated blood enters right atrium of heart via inferior vena cava and superior vena cava from systemic veins. The right ventricle of heart receives de-oxygenated blood from right atrium and pumps it via the pulmonary artery to the lungs where carbon dioxide is released and oxygen is received into the blood. The blood moves from the lungs via the pulmonary vein to the left atrium of the heart. Valves open and close at the entry to, between, and exit from, the atria and ventricles. The left atrium passes oxygenated blood to the left ventricle, which pumps the oxygenated blood out the large artery called the aorta. The aorta connects by systemic arteries to cerebral, coronary, renal, visceral (splanchnic), and skin vasculatures and to vasculature of skeletal muscles. The names of the valves are: tricuspid valve—right atrium to right ventricle; pulmonary valve—right ventricle to pulmonary artery; mitral valve—left atrium to left ventricle; and aortic valve—left ventricle to aorta.

The primary heart sound components, S1 and S2, are composite signals generated by valve closures. S1 is caused by the closure of the mitral and tricuspid values of the heart, and S2 is caused by the closing of the aortic and pulmonary valves. An analog electrical heart monitoring signal is captured by two or more ECG electrodes, and the signal is a varying voltage representing electrical activity of the heart, i.e., the signal generated in a person\'s body to cause the heart to contract or relax. The ECG signal has three main components, a P-wave, a QRS complex made up of a Q-wave, an R-wave, and an S-wave, and a T-wave. The pulses include a small positive P pulse, a larger negative-going QRS depolarization pulse near in time to the S1 heart sound, and a large positive-going T pulse near in time to the S2 heart sound. The P-wave represents the depolarization (electrical activation) of the atria of the heart. The QRS complex represents the ventricular activity of the heart. The T-wave represents the re-polarization of the ventricles.

Process and structure embodiments can also be extended to other biomedical signals corrupted by motion wander—e.g., ECG electrocardiogram, PPG—photoplethysmogram (signal from a Pulse Oximeter), EEG—electroencephalogram, EMG—electromyogram, ICG—Impedance Cardiogram signals—or almost any other signal that might be affected by a separable wander. Thus, motion-related artifacts are removed from such other biomedical signals in products that can be produced by a manufacturer in volume.

Remarkably, with some of the embodiments of structure and process, polynomial smoothing and differentiating functions and operations are performed. A secondary reference sensor or signal source is unnecessary. Gross motion is tracked and canceled out from the primary accelerometer-based signal. A polynomial smoothing filter 130 (for example, a Savitzky-Golay filter) is electronically instantiated herein and digitally smoothes a given accelerometer-based data signal stream by approximating it within a specified data window by a polynomial of a specified order that best matches the data in the window in a least-squares sense. Here, the electronic smoothing filter 130 fits the slower variations in body-motion-induced components of the biomedical sensor-based signal and subtracts them as smoothed content from the biomedical sensor-based signal to leave behind what is called a residue signal. The residue signal provides a thus-extracted, faster-varying signal—primarily the heart sounds and other cardiac activity, as well as some residual or remaining noise.

Such polynomial filtering 130 preserves higher order moments around inflection points, or at extrema like peaks and troughs, that a digital moving average or low-pass filter does not. In other words, the polynomial filtering better preserves features—like local maxima and minima—through a least-squares polynomial fit around each point. Also, unlike a moving average, in estimating the value of the fit at a certain point, it does not factor in the values on the polynomial fit around it, therefore not introducing a bias at such features while reducing the noise.

In FIG. 1, a system embodiment has hardware that provides a measurement set-up and monitoring embodiment. A miniature (weight—0.08 gram, size—5×5×1.6 mm) triple axis, low-power, analog output MEMS accelerometer (LIS3L02AL, STMicroelectronics, Geneva, Switzerland) is taped onto the chest (e.g., a few inches to the left of the sternum along the third or fourth rib). (Taping the accelerometer sensor or using a chest band presses the accelerometer sensor to or against a bare or shaved portion of the chest and efficiently couples chest acceleration to the sensor.) An acceleration signal corresponding to the cardiac activity is captured along the Z-axis—the dorso-ventral direction orthogonal to the plane of the chest. The chest acceleration signal is AC coupled with a 3 Hz cut-off and amplified with a gain of 100 and low pass filtered—for anti-aliasing—through a three-stage, 5-pole Sallen-and-Key Butterworth filter with a 1 kHz corner frequency. A commercial quad operational amplifier (op amp) package (LT1014CN, Linear Technology, Milpitas, Calif.) is used for the analog front-end. The accelerometer signal is then sampled at 10,000 Samples/sec using a data acquisition card (National Instruments, Austin, Tex.) and captured and stored on a computer using MATLAB software (Version 2007b, The Mathworks, Natick, Mass.).

The AC coupling with approximately 3 Hz cutoff, which is a non-critical rolloff frequency, is provided, for example, by a series coupling capacitor C coupled to an input resistance established for the amplifier.

In FIG. 2, a reference ECG (lead II) is acquired simultaneously in a three electrode (single lead) electrocardiogram ECG amplifier configuration as a standard of reference in order to compare with the accelerometer-derived cardiac signal for the evaluation of the performances of the heart rate extraction from the accelerometer signal.

In FIGS. 3 and 4, for detection of primary heart sounds and cardiac activity, the acceleration signal is digitally low pass filtered in a step 110 at 50 Hz—using a 3326 tap digital FIR filter with a steep 80 dB roll-off over 20 Hz—and decimated in a step 120 by a factor of 10. (Rolloff frequency less than 60 Hertz attenuates 60 cycle USA power line interference with biomedical signals of interest, and rolloff may be made less than 50 Hertz for applicable countries using 50 Hertz. While the rolloff frequency could be made higher, this FIR filter also desirably attenuates white noise above the frequency range of the signals being monitored.) Also in a Phase 1, a high order Savitzky-Golay polynomial smoothing filter 130, using 28th order and 401 point frame, is used to capture the relatively slow-varying motion wander and leave out the more rapidly varying heart sound signal components. (Matlab syntax for such filter is g=sgolayfilt(X,28,401) where g is the filter output and X is a latest input column vector of 401 sample values of windowed data.) In a Phase 2, the smoothing filter 130 output is subtracted in a step 140 from the decimated LPF output to obtain heart sounds S1 and S2. A folded correlation process in a step 160 then enhances and strengthens the polynomial filtered S1/S2 peaks in the motion-removed acceleration signal. Such folded correlation process 160 is described in further detail elsewhere herein and with background in U.S. patent application Publication “Heart Rate Detection In High Noise Conditions” 20100094150, dated Apr. 15, 2010 (TI-66732), which is incorporated herein by reference. Then the location of the peaks is threshold-detected in a step 170 using an electronic amplitude-based peak picking process, and the selected peaks are counted in a step 180 to calculate heart rate HR.

In FIG. 5, a chest-acceleration signal is derived from the accelerometer sensor while a subject is walking around a room and low pass filtered at 50 Hz (step 110) as shown in a first waveform. LPFing (low pass filtering) sub-50 Hz is used in some of the examples because most of the desired signal power lies in that range and in general LPFing with some rolloff frequency below about one hundred Hertz in many of the embodiments avoids making the bandwidth so wide as to encompass and integrate a substantial or undue amount of sensor noise (thermal, white spectrum). In case LPF with a rolloff frequency above power-line frequency is used, then some embodiments also include notch-filtering for power-line frequency rejection. In FIG. 5, a second waveform is an electronically-derived polynomial smoothing filter 130 output corresponding primarily to the body motion. A third waveform concurrently shows the residue signal after subtraction 140 in FIG. 4 and isolates the primary heart sounds. A simultaneous ECG timing signal is shown as a fourth concurrent waveform for reference.

In FIG. 6, the same embodiment monitors a chest-acceleration signal from the accelerometer Z-axis sensor while the subject walks on a treadmill. A brief rest recording is followed by motion. Compared to FIG. 5, the plot of FIG. 6 analogously shows an unfiltered (raw) acceleration signal and the residue from step 140 after the polynomial smoothing of step 130. Note the magnified scale in some parts of FIG. 6, and that FIG. 6 has a different scale than in FIG. 5.

FIGS. 7A-7C show signal plots for an ECG filtering embodiment 2. The plots have different time scales and walking conditions. Raw ECG signal from the ECG electrodes in FIG. 2 and a concurrent filtered ECG signal waveform, by applying steps 110-140 separately to the ECG signal, are depicted for a subject walking on a treadmill.

In another embodiment, satisfactory S1-S2 heart signals were extracted from raw motion-affected accelerometer Z-axis data by LPF (low pass filtering) with corner at 100 Hz and then Savitzky-Golay filtering at 20th order, followed by subtraction of the S-G signal from the LPF signal, and followed further by signal enhancement. It appears that polynomial filtering of motion-affected LPF accelerometer signals, using polynomial filtering on the order in a range of approximately 20th order or higher order to at least over 30th order, is satisfactory for obtaining heart signals as a residue by subtraction of the polynomial filtering output from the LPF signals. Using polynomial fits at such orders successfully captures both coarser and finer motion effects. The smoothing filter in some embodiments can be lower order as well, and may obtain good results even with a 1st-order polynomial in case of some window sizes and applications. Also, lower order polynomial filtering is contemplated and found useful as discussed later hereinbelow. Using a number of points at least approximately half again (1.5 or more times) an order of the polynomial and even substantially higher than that, in some of the embodiments, is believed to help to reduce noise.

In FIG. 8, a wireless embodiment has the accelerometer sensor 210 in a chest-worn miniature unit including Bluetooth or pico-network wireless or an RF transponder. The miniature unit 210 wirelessly communicates with a Data acquisition/signal processing unit 215, 220 of FIGS. 8 and 4 such as provided on a belt clip, in a cell phone or in a gateway elsewhere in a residence (see FIG. 39). In FIG. 8, the signal processing unit 220 is coupled to a wireless modem 230 or transmitter (or to a wireline modem) for transmission to a remote location such as a medical clinic. The medical clinic has a receiver or transceiver such as in a cell phone or wireless or wireline modem 240, and further has a data storage and display unit 250. The medical clinic can interrogate the residential data acquisition/signal processing unit 220 by transceiver 240 via residence-based modem 230 and re-configure the residential unit 215, 220 for various performances and for more or less information and more or less frequent communications.

In FIGS. 3 and 8, any two or more, or all, of the described components can combined in a single digital system. The monitoring signal capture component 210 is configured to capture a heart monitoring signal from a person and provide it to an analog signal conditioning and sampling section 215 (A-to-D) that feeds digital data to the signal processing component 220. In some forms, the A-to-D happens physically within the accelerometer chip and the signal flow remains electronically arranged as shown. Either or both of components 210 and 215, 220 may provide amplification and noise reduction of the analog and/or digital signal in the process. In various embodiments, the heart monitoring signal may be provided to the processing component 220 in real-time, and/or may be provided periodically as the signal is being captured, and/or may be recorded and provided to the processing component at a later time.

The digital heart monitoring signal may be provided to the data acquisition 215 and signal processing unit 220 by wired or wireless forms of communication, e.g., wired using a USB port, electrode wires, logic circuitry, etc. or wirelessly such as by a Bluetooth connection, Zigbee, or otherwise. In FIG. 8 a communications network for remote transmission can be wide area network (WAN) such as the Internet, a wireless network, a local area network (LAN), or a combination of networks. Similarly, the processing component 220 may be connected to an output component 250 by any of the foregoing connections and networks. Any suitable display device and/or recording apparatus 250 is used such as, for example, a computer monitor, a display of a handheld computing device, a display in a personal heart rate monitoring device, a display in a piece of exercise equipment, etc. The system hardware of FIG. 39 may be applied with one program at both the premises at which the accelerometer is used and a replica of that system hardware applied with that program and/or an additional program at the remote premises such as a medical clinic.

The system components including signal processing component 220 may also be implemented by or as part of any suitable digital system (e.g., a general purpose processor, a digital processor, a personal heart rate monitoring system, a heart rate monitoring system in a piece of exercise equipment, a personal computer, a laptop computer, a server, a mainframe, a personal digital assistant, a television, a cellular telephone, an iPod, an MP3 player, etc.) configured to receive the digital heart monitoring signal from the monitoring signal capture component 210. The processing component 220 is configured to process the digital heart monitoring samples in the digital heart monitoring signal in accordance with embodiments of methods described herein. In one or more embodiments of the invention, the processing component 220 includes functionality, e.g., a computer readable medium such as memory, a flash memory, an optical storage device, a disk drive, flash drive, etc., to store executable instructions implementing an embodiment of a method for processing heart monitoring samples as described herein and to execute those instructions.

Embodiments like those of FIGS. 8 and 4 and other Figures herein have potentially wide-ranging applications from commercial products that already have in-built accelerometers (e.g., mobile phones, personal entertainment devices, content players, computer game controllers etc.) and those that do not (clothing, accessories etc.) to fitness products (heart straps, belts, wearable adhesive bandages or sensor tapes, clips, straps, bands or carriers for temporary affixing to one\'s chest, arm or elsewhere on the body, or implantable sensor devices) Embodiments are suitably made as a part or whole of ambulatory monitoring products for ambulances, at trauma sites (e.g., for accident or burn victims), for home-monitoring of older adults and all populations to which the advantages of the embodiments commend themselves.

The accelerometer 210 signals from all three axes are suitably also processed to electronically double-integrate the acceleration to determine the location of the person wearing it. Since the person is likely to have been in bed overnight, the processing determines the location of the person during the day by double-integrating the acceleration starting from initial conditions of position initially at the bed location, and zero initial vector velocity. This information can be helpful as a cue to the person who is visually impaired, to care-giver, and to a family member. The accelerometer processing can indicate that the person is in a given room of the residence, as an assist for one who is visually impaired, or can indicate that the person is leaving or has left the residence to inform a care-giver or family member. In this way, the accelerometer and associated processor provide numerous services for all concerned, in various ways as taught herein.

For background on accelerometer calibration and double-integration see U.S. patent application “Parameter Estimation for Accelerometers, Processes, Circuits, Devices and Systems” Ser. No. 12/398,775 (TI-65353) filed Mar. 5, 2009, which is incorporated herein by reference in its entirety.

Due to its low-cost and ease of use, products using the embodiments have potential for commercial success not only in urban and developed areas but also widely in the developing world as well as in rural parts of the developed world or in any place where low-cost, remote health monitoring facilities may be rare, if available at all.

The smoothing filter 130 of FIG. 4 is configured based on a specified order M and frame size (number of sample points N). For instance, a Savitzky-Golay polynomial smoothing filter is used in some embodiments to best approximate the acceleration signal in the least-squares sense to capture the motion-dependent baseline-wander. In some embodiments, the smoothing filter is implemented in flash memory of a local processor of FIG. 39 such as a belt-worn unit or provided in a home network gateway or clinic office network gateway, or cell phone or otherwise.

The matter of selecting and or finding feasible and optimum values for order M and window length (NW in points, tW in time) for the polynomial smoothing filtering is discussed next. In general for a fixed window length, NW, a higher order polynomial will fit the high frequency components of the streaming data better. For a given order M, a shorter window of time will allow fitting the high frequency component better.

In FIG. 4, the working hypothesis is that the accelerometer signal has a low-frequency (motion) component and a high frequency (heart signal) component. The polynomial filter is used to fit to the motion component.

A way to approach the optimization problem estimates the inherent order of the low-frequency component and picks the smallest window that satisfies the condition that NW>M+1 and NW is odd (i.e., NW=2N+1). The smaller window size NW is, the smaller is the number of taps of the multiply-accumulate filter process implementing the smoothing filtering. For an accelerometer signal in some applications, order M=1 and window size NW=3 (sampling frequency is 1000 Hz). In some examples herein, higher orders M and window widths NW are shown.

In FIG. 4, motion, heart sounds and heart rate are electronically separated and ascertained from accelerometer 210 data using the following steps:

a) Low-pass filtering 110 and decimating 120 the accelerometer data b) Savitzky-Golay filtering 130 to fit the relatively lower frequency motion data c) Subtracting 140 the output of the Savitzky-Golay filter from the low-pass filtered accelerometer data (from step a) to obtain the heart sounds d) Performing 160 folded correlation to enhance the primary heart sounds (S1 and S2) peak locations e) Peak picking 170 to count the number of S1 peaks in a predetermined or configured segment (time interval) and counting 180 the heart rate HR in beats per minute BPM.

Note that the term ‘decimation’ refers to any process of regularly removing samples from a sample stream, or passing one sample in every nD samples as decimation parameter, and can but does not necessarily refer to removing all but 1 sample in ten. Thus, if a sample/ADC delivers fS samples per second, then a decimation process delivers a decimation frequency substantially fS/nD samples per second. If a window period is tW seconds, then the number of points NW=2N+1 in the window is NW=1+fS tW/nD. The window period tW may be selected by considering the time period over which the particular features and behavior of interest are to be obtained by the filtering from the signal. The sampling frequency fS may be selected with cost, physical size and complexity of anti-aliasing in mind (low pass filter AAF at 0.5fS or less situated ahead of sampling fS). The sampling frequency fS may be set substantially greater than the Nyquist frequency for sampling the AAF output. The decimation parameter nD is then selected, firstly, to yield a decimation frequency fS/nD that is sufficiently high relative to the e.g., 50 Hz low pass filter LPF following the sampling/ADC circuit to provide effective operation of that LPF. Secondly, the decimation parameter nD is also selected to yield a number NW of window points that is sufficiently high relative to the selected order M of the filter to keep filter noise low while having the NW window points being sufficiently low in number as to introduce only so many filter computations as needed to achieve satisfactory filtering of the signal stream in the window. The filter computations are related to the product of the number NW of points per window multiplied by a rate number rW of windows processed per second. If rW=NW/tW, the computations are proportional to NW2/tW, which may motivate fewer window points and longer window times in some energy-saving and lower cost processor applications. Remarkably, the examples herein satisfy these considerations for some applications and other examples may readily be devised for other particular applications as well.

Mathematically expressed processes are described in further detail below for preparing various electronic embodiments with smoothing filters for various ways of motion extraction in step 130 and any other purpose to which their advantages commend their use. They are appropriately partitioned into offline and real-time online electronic processes in such embodiments.

The notation ∥(x−Ab)∥ in Equation (1) signifies the sum of squared differences between the [(2N+1)×1] respective data stream vector sample points or stream components and the (2N+1) respective estimates of those stream components provided by multiplying a [(2N+1)×M] transform matrix A times a [M×1] vector of transform coefficients bj. The number of transform coefficients bj is M, and they form a [M×1] vector b. A gradient V is the [M×1] vector of first partial derivatives with respect to the transform coefficients bj. The number M of coefficients bj is called the order, and if the number of transform coefficients bj is M, then the order of the process is M. The [M×M] matrix of second partial derivatives with respect to the transform coefficients bj is signified by ∇∇. The filter procedure involves, and in effect forms, a coefficients change coefficient vector Δb for updating an initial transform coefficient estimate b=0 (i.e., all coefficients initialized to zero). This procedure pre-multiplies the matrix of second partial derivatives times the negative of the gradient to obtain that transform coefficients change vector Δb.

Δb=−(∇∇∥(X−Ab)∥)−1∇∥(X−Ab)∥  (1)

Since the Equation (1) involves a quadratic expression and starts from b=0, the process directly finds the values of the transform coefficients b=Δb in one pass without iterating additionally. Equation (2) represents the result of performing the calculus operations represented by Equation (1). (Some embodiments transmit the coefficients b from Equation (2) to a remote site for record storage and further analysis, since they effectively compress much of the information in the data window. If coefficients are to be transmitted, the [M×(2N+1)] matrix (ATA)−1 AT is pre-computed and then multiplied by each data window locally on the fly. Other embodiments omit such compression and/or transmission, or only do it locally on remote command, and thereby save some power and processing complexity.)

b=(ATA)−1ATX  (2)

This process generally finds transform coefficients bj provided the inverse (ATA)−1 exists. That inverse exists when the rows of the matrix A are linearly independent (full rank) and enough data points NW=(2N+1) are provided so that the corresponding number of columns of the matrix is sufficient for an inverse to be delivered.

In the special case of a polynomial transform process, a matrix of indices is raised to powers, wherein the jth column element Anj in the nth row of transform matrix A is raised to a power: nj. In other words, for the 2N+1 different values of n from −N to +N in the window of a data stream X(i+n), the transform finds a set of coefficients bj for a well-fitting power series to approximate all the values. Such a power series in general is represented by Equation (3):

X′(i+n)=bo+b1n+b2n2+b3n3+b4n4+ . . . bMnM  (3)

Savitzky-Golay filtering outputs as the filter output g(i) for the window indexed i the value of b0 estimated by Equation (2) for each data window, and successively window-by-window for successive indices g(i).

Rows of matrix A are orthogonal when the inner product is zero for any pair of different ones of them. These rows are illustrated in TABLE 1. The rows of values Anj in matrix row n are non-orthogonal for the example of a polynomial transform. (“̂A” signifies raising to a power.)

TABLE 1 ARRANGEMENT OF MATRIX AT Power m (Order M) Points 0 1 2 3 . . . M n = −N: [1 (−N) (−N){circumflex over ( )}2 (−N){circumflex over ( )}3 . . . (−N){circumflex over ( )}M]. . . . n = −4: [1 −4 (−4){circumflex over ( )}2 (−4){circumflex over ( )}3 . . . (−4){circumflex over ( )}M] n = −3: [1 −3 (−3){circumflex over ( )}2 (−3){circumflex over ( )}3 . . . (−3){circumflex over ( )}M] n = −2: [1 −2 (−2){circumflex over ( )}2 (−2){circumflex over ( )}3 . . . (−2){circumflex over ( )}M]

Download full PDF for full patent description/claims.




You can also Monitor Keywords and Search for tracking patents relating to this Estimation of blood flow and hemodynamic parameters from a single chest-worn sensor, and other circuits, devices and processes patent application.
###