Automated seismic detection of myocardial ischemia and related measurement of cardiac output parameters

• US 6,024,705 A
• Filed: 11/06/1998
• Issued: 02/15/2000
• Est. Priority Date: 11/06/1998
• Status: Expired due to Fees

• Current Assignee: Bioacoustics, Inc.
• Original Assignee: Bioacoustics, Inc.
• Inventors: Boehlen, Bruce H., Gorski, Stephen H., Schlager, Kenneth J.
• Primary Examiner(s): Getzow, Scott M.
• Application Number:
  US09/188,069
• Time in Patent Office: 466 Days
• Field of Search: 600/508, 600/509, 600/510, 600/511, 600/512, 600/513, 600/514, 600/515, 600/528
• US Class Current: 600/508
• CPC Class Codes: A61B 5/1102 Ballistocardiography
• CPC Class Codes: A61B 5/7203 for noise prevention, reduction or removal
• CPC Class Codes: A61B 5/7264 Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems neural networks per se G06N3/00; expert systems per se G06N5/00
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

A computer-based instrument to produce a "number" for heart performance parameters and a positive-negative diagnosis of myocardial ischemia. A seismic sensor captures a substantial series of SCG waveforms within a short time frame. Digitized waveforms are created and processed to create signals in the range of 0 to 50 hertz and 0 to 100 hertz. The waveform are processed in the time domain. The 0 to 100 hertz signal is processed to determine the heart rate which is pulse adjusted and interpolated. The SCG waveforms are processed to synchronize the start point of each waveform. The 0 to 50 hertz signal is then processed for signal segmentation to produce waveform signals, each a heart beat or period in length. The segmented signals are then processed to produce linear prediction analysis (LPA) coefficients. The coefficients establish a numerical model-based representation of the waveform. The LPA coefficients in combination contain all of the information resident in the original SCG waveform. For myocardial ischemia analysis, proper LPA coefficients are used in a pattern recognition algorithm to determine a classification of the patent'"'"'s waveforms as either normal or ischemic. The Bayesian decision classifier provides an analytical framework and program for classification of SCG waveforms as represented by the LPA coefficients for myocardial ischemia, or other cardiac disease conditions represented in the SCG waveform, and produces a direct negative or positive output. For various cardiac performance parameters, estimation rather than a classification algorithm is used such as a K-Nearest Neighbor pattern recognition technology, and multiple regression estimators and produces estimation for different parameters.

Claims

1. A method of rapid, noninvasive automated detection of myocardial ischemia secondary to coronary artery disease in a patient, comprising:

connecting a seismic sensor to the patient and detecting a series of SCG waveforms in accordance with the patient'"'"'s heart functioning, converting each said SCG waveform into a digital waveform in the time domain, and generating a list of coefficients for said digital waveform containing the complete information resident in said waveform.

2. The method of claim 1 including generating a set of model coefficients of a model SCG digital waveform;

comparing said coefficients of said SCG waveform of the patent with said model coefficients to product an output related to said comparison.

3. The method of claim 2 wherein said comparing step includes a computerized pattern recognition program.

4. A method of rapid, noninvasive automated detection of myocardial ischemia in a patient, comprising:

connecting a seismic sensor to the patient and detecting a series of SCG waveforms in accordance with the patient'"'"'s heart functioning, converting each said SCG waveform into a digital waveform in the time domain, and generating a list of coefficients for said digital waveform containing the complete information resident in said waveform,

processing each said filtered digital waveform by segmenting said digital waveform into a substantial plurality of segments, and generating a set of linear prediction coefficients for each segment to define each waveform with the total information resident in the SCG waveform,

comparing said defined waveforms using a Bayesian process with model reference SCG waveforms representative of both non-ischemic conditions and ischemic conditions for the presence of an ischemia condition and creating a direct status output for the patient, and presenting said output.

5. The method of claim 4 including the step of processing said linear prediction coefficients with a nearest neighbor program and form multi-neighbor clusters for each said defined waveform.

6. The method of claim 5 including a probability of accuracy.

7. The method of claim 4 wherein said output is a single positive or negative output.

8. The method of measurement of a heart performance parameter of a patient comprising connecting a seismic sensor to the patient and generating a series of SCG waveforms in accordance with the patient heart functioning,

converting each said SCG waveform into a digital waveform,

filtering said digital waveforms to a frequency of 0-50 Hz.,

processing each said digital waveform by segmenting said digital waveform into a substantial plurality of segments and generating a set of linear prediction coefficients for each segment to define each waveform with the total information resident in the SCG waveform, and employing a selected subset of these coefficients in a statistical estimation process to estimate cardiac performance parameters.

9. The method of claim 8 wherein said statistics estimation is a K-nearest neighbor process.

10. The method of claim 8 wherein said statistics estimation is a linear regression process.

11. The method of claim 8 wherein said parameter is selected from a plurality of different known heart parameters.

12. The method of claim 8 including the step of processing said linear prediction coefficient with a K nearest neighbor program and form multi-neighbor clusters for each said defined waveforms.

13. The method of detecting a varying and cyclical occurring condition represented by a corresponding continuous complex waveform including a condition complex waveform for each cycle based upon a reference condition complex waveform, comprising generating a selected sequence of condition complex waveforms in said continuous complex waveform, segmenting each said condition complex waveform in said sequence into a substantial number of corresponding segments, developing a set of linear prediction coefficients for each said segment for each said waveform and thereby presenting each said waveform as a series of digital subsignals and presenting a corresponding digitized subsignal containing all of the information of complex waveform, and comparing in a pattern recognition system each said digitized subsignals with reference condition complex waveforms representative of a known condition reference state of said varying and cyclically occurring condition for classifying and varying of said condition with respect to said reference state and cyclically occurring condition.

14. The method of claim 13 wherein a Burg algorithm calculates an L-matrix as a base for selecting said linear prediction coefficient, analyzing said L-matrix for selecting the preferred best linear prediction coefficients in combination for said segments.

15. The method of claim 14 wherein said signal comparing includes the Bayesian parametric system for processing the coefficients of each segment of the digitized signal with said references includes a Bayesian classifier program.

16. The method of claim 15 including a nearest neighbor estimation program to convert said digitized signals for classifying said condition including a possible heterogeneous character of the possible varying conditions.

17. The method of claim 13 wherein said condition waveform is an SCG waveform signal.

18. The method of detecting a varying and cyclically occurring condition represented by a continuous complex waveform of said condition including a plurality of like condition complex waveforms, comprising:

generating a sequence of said condition complex waveforms, segmenting each said continuous condition complex waveform into a substantial number of corresponding and like segments, development a set of linear prediction coefficients for each said segment for each said condition complex waveform whereby each said waveform includes a series of digital subsignals and thereby creating corresponding digitized subsignals containing all of the information of the continuos waveform, establishing a pattern recognition system in which each said digitized subsignal is processed with reference waveforms representative of a known reference condition state of said continuous complex waveform including the varying cyclical occurring condition.

19. The method of claim 18 wherein said developing said set of linear prediction coefficients for each said segment processes a Burg algorithm including calculating an L-matrix as a base for selecting said linear prediction coefficients, and analyzing said L-matrix and thereby selecting the preferred and optimal linear prediction coefficients in combination for each of said segments.

20. The method of claim 19 wherein said comparing and execution of said pattern recognition system includes a Bayesian parametric system for processing and analyzing the coefficients of each segment of the digitized signals with said reference waveforms including execution of a Bayesian classifier program.

21. The method of claim 19 wherein the analyzing of said L-matrix includes a nearest neighbor estimation program for processing said digitized signals to produce a condition signal related to a heterogeneous character of the cyclically occurring condition.

22. The method of claim 18 wherein said condition signal is an SCG waveform signal.

23. The method of claim 18 including providing a seismic sensor to detect the varying and cyclically occurring condition and thereby generating said sequence of said condition complex waveforms,

converting each of said SCG waveforms into a digital waveform, filtering said frequency of interest with relation to a condition to be monitored or analyzed, synchronizing the plurality of waveforms to establish a common starting point and each waveform for analysis purposes, determining the original condition rate creating said series of waveforms, normalizing the condition rate to normalize the condition rate represented in said complex waveform and thereafter including said segmenting step and final processing.

24. A method of non-invasive detection of the heart of a patient relative to a known class of patients;

comprising;

generating an SCG waveform corresponding to the patient'"'"'s heart functioning,

converting each of said SCG waveforms into a digital waveform,

filtering each of said digital waveforms to a first output requency of 0-100 Hz. and to a second output frequency of 0-50 Hz.,

processing the 0-100 Hz. filtered SCG signal in a unipolar autocorrelation function to determine the repetitive heart rate,

synchronizing of the starting point of each SCG waveform for subsequent processing,

processing the 0-50 Hz. output frequency signal to segment the signal into a series of segments one heart beat in length and developing the LPA coefficients including the information resident in the original SCG waveforms, and

comparing said coefficients with coefficients of the waveform.

25. The method of claim 24 including the step of determining the mean and standard deviation of each linear prediction coefficient and rejecting coefficients outside of at least two signal standard deviation range from further processing.

26. A computer based instrument for the noninvasive detection of the heart condition of a patient relative to a known class of patients, said known class of patient including heart waveforms identified by a set of linear prediction coefficients, comprising:

a signal source configured to sense and generate an SCG waveform corresponding to the patient'"'"'s heart functioning for a selected period of time, a converter connected to said signal source for converting said SCG waveform into a digital signal,

a computer connected to said converter, said computer including;

an input connected to said converter,

a filter program for filtering said digital signal to an output signal of a frequency of 0-50 Hz.,

a program developing LPA coefficients including the information resident in the original SCG waveform of the patient based on said 0-50 Hz. digital signal, and

a program comparing said coefficients of said patient'"'"'s signal waveform with the coefficients of said known class of patients waveform for determining the heart functioning of the patient.

27. The instrument of claim 26 wherein said computer includes a Bayesian classifier program processing of said LPA coefficients of the waveform for ischemic heart status, said Bayesian classifier executing such comparing program and produces a negative or positive or output with respect to the condition of said ischemic condition.

28. The computer based instrument of claim 26 wherein said computer includes a program inclusive of heart waveforms for said known class of patients related to one or more parameters of the heart functioning, said program developing said LPA coefficients developed programs or LPA coefficients related to a selected parameter, and said comparing program determines an estimation of the heart functioning of the patient.

29. The computer based instrument of claim 26 wherein said computer includes a synchronizing program for synchronizing the starting point of the individual heart beat waveform resident in said SCG waveform for processing of said 0-50 Hz. frequency signal.

30. The instrument of claim 29 wherein said computer includes a program for determining the means and standard deviation of each of said linear prediction coefficients and rejecting coefficients outside of at least two standard deviation ranges from final processing and the determination of the execution of the comparing program.

31. The instrument of claim 26 wherein said computer includes a computer program establishing an output frequency of 0-100 Hz., and includes a program executing a heart rate determination program based on said 0-100 Hz. signal, and a program processing said 0-50 Hz. signal to segment the waveform into a series of segments each being one heart beat in length.

32. The instrument of claim 31 including a phase synchronizing program to synchronize the heart signal waveforms within said SCG waveform, said program developing said LPA coefficients being based on said segments of said SCG waveform.

33. The instrument of claim 32 wherein said computer includes a sample rate determination program, said computer executing said sample rate determination program after determination of the heart rate and prior to the execution of said segmenting of the signal into a series of segments.

34. The instrument of claim 32 wherein said computer includes an SCG screening program to screen out premature ventricular contractions and left bundle branch block of the SCG waveform prior to the comparing program.

35. The instrument of claim 26 wherein said LPA related coefficient determination program develops coefficients related to a particular parameter of the heart functioning, said comparing program including a K-Nearest neighbor estimator including coefficients relating to said parameter, and develops and produces an estimation of the actual heart functioning based on the comparison of the known and processed coefficients.

36. The instrument of claim 26 wherein said LPA related coefficient determination program develops coefficients related to a particular parameter of the heart functioning, said comparing program includes a linear regression program for developing the coefficients for a parameter, and said computer including a multiple regression estimation estimator for producing an estimation of the heart parameter functioning.

37. The instrument of claim 26 wherein said LPA related coefficient determination program develops coefficients related to a particular parameter of the heart functioning, said comparing program including a systolic time interval program for detecting or including a resident known coefficients for MC-AO function and the LVET function of a heart, and providing an output related between the known coefficient and the developed coefficients, and a multiple regression estimator providing an estimation output of the actual heart functioning of the patient.

38. The instrument of claim 26 wherein said comparing program generates an output related to the ejection fraction of the heart and the stroke volume of the heart, and includes program for developing of other heart parameters based on said outputs and the heart rate.