Method and apparatus for spectral analysis of electrocardiographic signals

• US 5,109,862 A
• Filed: 02/21/1991
• Issued: 05/05/1992
• Est. Priority Date: 03/19/1990
• Status: Expired due to Term

• Current Assignee: DEL Mar Medical Systems, LLC (Bio-Techne Corporation)
• Original Assignee: Del Mar Avionics
• Inventors: Henkin, Raphael, Kelen, George J.
• Primary Examiner(s): Kamm, William E.
• Application Number:
  US07/658,505
• Time in Patent Office: 439 Days
• Field of Search: 128/696, 128/702-705, 364/413.06
• US Class Current: 600/515
• CPC Class Codes: A61B 5/333 Recording apparatus specially adapted therefor
• CPC Class Codes: A61B 5/7253 characterised by using transforms
• CPC Class Codes: A61B 5/7257 using Fourier transforms

Abstract

A frequency domain signal processing and analysis method and apparatus displays, plots and makes measurements upon electrocardiographic signals (ECG) recorded from the body surface. Both the graphic plots and the numeric parameters measured reveal abnormalities of electrical conduction within the heart thought to be important in the identification of patients at risk of serious disturbances of heart rhythm including sudden death. The invention employs Fourier analysis of short overlapping segments of ECG signal to create a three dimensional map ("spectrocardiogram") whose axes are time, fequency and power thus disclosing changes in the frequency content of the ECG signal over short intervals of time. In additoin to the Fourier analysis itself, the method and apparatus provides a wide choice of user-selectable signal pre-processing, post-processing, display and plotting options as well as performing custom mathematical computations upon the Fourier spectra to distinguish and quantify visually apparent differences between normal and abnormal spectrocardiograms.

Claims

1. A method for detecting abnormalities in physiological functions of an organism by performing a frequency analysis of a single cycle of a time varying bioelectric signal, said method comprising:

(a) partitioning said single cycle of said time varying signal into a sequence of time-displaced, overlapping segments or slices,

(b) performing a frequency analysis of each of said slices, thereby resolving each of said slices into a sequence of discrete spectral components of varying amplitude,

(c) storing a number proportional to the magnitude of each of said spectral components in a table in which frequency is plotted along a first axis, and time is plotted along a second axis,

(d) assessing the degree of difference between the spectra of different slices of said table, and (e) comparing said spectral differences with corresponding differences obtained from representative samples of normal and abnormal test organisms.

2. The method of claim 1 in which said frequency analysis is further defined as a Fourier analysis.

3. The method of claim 2 wherein said Fourier analysis is further defined as a discrete fast Fourier transform (FFT).

4. The method of claim 3 wherein said discrete FFT is performed on overlapping segments of a selected portion of said signal.

5. The method of claim 4 further including the step of signal averaging said bioelectric signal prior to performing said FFT.

6. The method of claim 5 further including multiplying the amplitude of said signal by a windowing function prior to performing said FFT, thereby minimizing spectral leakage and effects of edge discontinuities inherent in said FFT process.

7. The method of claim 6 further including the step of displaying said spectral components on a contour plot approximating a three-dimensional plot in which frequency is plotted along one axis, the amplitude of each spectral component is plotted along a second axis, and time is plotted along a third axis, thereby facilitating a visual assessment of the degree of irregularity or spectral turbulence in said signal which may be associated with a physiological defect.

8. The method of claim 6 further including displaying numbers calculated from values of said spectral components.

9. The method of claim 4 further including the steps of calculating the mean value of the amplitude of each of said slices and subtracting said mean value from the instantaneous value of each said slice prior to performing said FFT.

10. A method for analyzing in the frequency domain a single cycle of an ECG signal, which single cycle may be an ensemble average of a plurality of cycles, said method comprising:

a. locating within said ECG signal, a time interval of interest, typically the QRS region of said ECG signal;

b. dividing said time interval of interest into a sequence of overlapping segments, or slices;

c. multiplying each said slice by a windowing function to form a windowed slice;

d. performing a discrete fast Fourier transform (FFT) on each said windowed slice, in which FFT the power spectral density (PSD) at a sequence of frequencies is calculated;

e. producing a numerical matrix or table of PSD'"'"'s in which discrete frequencies head each column in said table and discrete times head each row of said table, thereby resulting in a matrix of PSDs for a vertically-stacked array of time-slices, and,

f. comparing said matrix of PSD s with corresponding matrices obtained from representative samples of normal and abnormal test subjects.

11. The method of claim 10 further including calculating the mean value of the amplitude of each time slice prior to performing said FFT.

12. The method of claim 11 further including subtracting said mean value of each time slice from each data point within that time slice, thereby removing any D.C. offset in said time slice.

13. The method of claim 10 further including performing steps (a) through (f) on at least one additional ECG signal obtained from a patient simultaneously with said first ECG signal, thereby obtaining a plurality, N, of ECG signals and N PSD matrices.

14. The method of claim 13 further including calculating the mean values of each time slice of each ECG signal of said plurality of signals.

15. The method of claim 14 further including subtracting said mean value of each time slice of each of said ECG signals from each data point within that time slice, thereby removing any D.C. offset in said time slice.

16. The method of claim 14 further including summing said mean values of said N PSD matrices, thereby forming a (N+1)th average PSD matrix.

17. The method of claim 10 further including summing each PSD value in a time slice row to form a total PSD sum, thereby forming a total PSD column.

18. The method of claim 17 further including comparing each total PSD value in each vertical position in said PSD column with each other PSD value to determine that time slice row having the highest total PSD value, and designating said time slice row as the fiducial, or reference time slice of said QRS complex.

19. The method of claim 18 further including locating the onset and offset of the QRS portion of said ECG signal.

20. The method of claim 19 wherein said method of locating said onset and offset of said QRS portion of said ECG signal comprises:

(a) calculating the mean background noise value and standard deviation of said ECG signal represented by said total PSD matrix by identifying a given time interval, represented by a pre-determined number of consecutive time slices within said PSD matrix having the lowest average total PSD; and

(b) identifying that uppermost (earliest) row and lowermost (latest) row away from said QRS fiducial slice having a total PSD at least N1 standard deviations greater than said mean background noise.

21. The method of claim 20 further including calculating the sum of the total PSD'"'"'s over a terminal interval T1 of said QRS complex.

22. The method of claim 20 further including dividing said QRS region of said PSD matrix into a low power terminal region, and a high power region, said low power terminal region being defined as commencing at the first time slice following said fiducial QRS slice having a total PSD less than a percentage P1 of said fiducial slice, and ending at the termination of the QRS complex, and said high power region of said QRS region being defined as the remainder of said QRS region of said PSD matrix.

23. The method of claim 22 further including separate processing of said low power region data and said high power region data.

24. The method of claim 23 further including counting those peaks or maxima occurring along the frequency axis or rows of said PSD matrix table (frequency axis peaks) below a predetermined frequency F1 and storing for display the number LFP corresponding to said count, and counting those peaks or maxima occurring above said frequency F1 and storing for display the number HFP corresponding to said second count.

25. The method of claim 23 further including counting those peaks or maxima occurring along said time axis or columns (temporal axis peaks) and storing for display a number TP corresponding to said count.

26. The method of claim 25 further including subdividing for display said number TP into a low frequency count LTP for temporal axis peaks below a frequency F2 and a high frequency count HTP for temporal axis peaks above said frequency F2.

27. The method of claim 26 further including counting those peaks or maxima occurring simultaneously along the frequency axis or rows of said PSD matrix and along said time axis or columns of said PSD matrix, and storing for display a number TBP (total biaxial peaks) corresponding to said count.

28. The method of claim 27 further subdividing for display said number TBP into a low frequency count LBP for biaxial peaks below a frequency F3 and a high frequency count HBP for biaxial peaks above said frequency F3.

29. The method of claim 23 further including calculating for each lead'"'"'s PSD matrix table the Pearson correlation coefficient of each row with the row immediately below it (inter-slice correlation coefficient) and storing for display an additional column of numbers corresponding to said correlation coefficients.

30. The method of claim 29 further including counting the number LISC of said inter-slice correlation coefficients below a pre-determined number N2, and storing for display said number.

31. The method of claim 29 further including calculating and storing for display the mean of said inter-slice correlation coefficients (MISC) and standard deviation of said inter-slice correlation coefficients (SDIC).

32. The method of claim 31 further including calculating and storing for display a number SKISC corresponding to the skew of the distribution of said inter-slice correlation coefficients by dividing the number of instances of said inter-slice correlation coefficient being less than said mean value MISC, by the total number of said slices.

33. The method of claim 29 further including calculating and storing for display a number called spectral entropy by adding an additional row to said PSD matrix table, said additional row containing the calculated average PSD at each harmonic frequency of selected rows of said PSD, calculating the average correlation of each time slice row with the new row of average PSDs, and subtracting said average correlation from 1.

34. An apparatus for analyzing ECG signals to detect abnormalities of electrical conduction within the heart comprising:

(a) means for selectably storing a waveform representative of the amplitude-versus-time variation of a single cycle of said ECG signal, or an ensemble average of said cycles;

(b) means for locating within said waveform a time interval of interest;

(c) means for dividing said time interval into a sequence of overlapping segments, or time slices;

(d) means for performing a frequency analysis of each of said overlapping segments;

(e) means for producing a numerical matrix table of the amplitude of each frequency component obtained by said frequency analysis means, said matrix table having in a first line of headings discrete frequency values, and in a second orthogonal line of headings, discrete time intervals corresponding to the beginning of each slice, and

(f) means for comparing said table with corresponding tables obtained from normal and abnormal test subjects.

35. The apparatus of claim 34 wherein said means for performing said frequency analysis is further defined as effective in determining the relative amplitude of discrete Fourier components, or power spectral density components (PSD'"'"'s), into which said waveform may be resolved.

36. The apparatus of claim 35 further including means for calculating the mean value of the time-domain amplitude of each of said time slices, prior to performing said frequency analysis.

37. The apparatus of claim 36 further including means for subtracting said mean time-domain value of each time slice from each data point value within that slice, thereby removing any D.C. offset in said time slice.

38. The apparatus of claim 36 further including means for summing each PSD value in a time slice row to form a total PSD sum, thereby forming a total PSD column.

39. The apparatus of claim 38 further including means for comparing the total PSD value in each vertical position in said PSD column with each other PSD value and determining that time slice row having the highest total PSD value, thereby determining a fiducial or reference time slice of the QRS complex of said ECG waveform.

40. The apparatus of claim 39 further including means for locating the onset and offset of said QRS portion of said ECG signal.

41. The apparatus of claim 40 further including means for calculating the sum of the total PSD'"'"'s over a terminal interval T1 of said QRS complex.

42. The apparatus of claim 41 further including means for segmenting said QRS region of said PSD matrix into a low power terminal region and a high power region, said low power terminal region being defined as commencing at the first time slice following said QRS slice having a total PSD less than a percentage P1 of said fiducial slice, and ending at the end of said matrix, and said high power portion of said QRS region being defined as the remainder of said QRS portion of said PSD matrix.

43. The apparatus of claim 42 wherein said further including means for further separate processing of said low power terminal region and said high power region, said means for further processing (as defined as) including means for counting those peaks or maxima occurring along the frequency axis rows of said PSD matrix table (frequency axis peaks), those peaks or maxima occurring along the time axis or columns of said PSD matrix table (temporal axis peaks), and those peaks occurring in both axis simultaneously (biaxial peaks).

44. The apparatus of claim 43 further including means for calculating for each said PSD matrix table the Pearson correlation coefficient of each row with the row immediately below it (inter-slice correlation coefficient) and storing for display an additional column of numbers corresponding to said correlation coefficients.

45. The apparatus of claim 44 further including means for counting the number LISC of said inter-slice correlation coefficients below a pre-determined number N2, and storing for display said number.

46. The apparatus of claim 45 further including means for calculating and storing for display the mean of said inter-slice correlation coefficients (MISC) and the standard deviation of said inter-slice correlation coefficients (SDIC).

47. The apparatus of claim 46 further including means for calculating and storing for display a number SKISC corresponding to the skew of the distribution of said inter-slice correlation coefficients by dividing the number of instances of said inter-slice correlation coefficients being less than said mean value, by the total number of said slices.

48. The apparatus of claim 44 further including means for calculating and storing for display a number called spectral entropy by adding an additional row to said PSD matrix table, said additional row containing the calculated average PSD of each harmonic frequency of selected rows of said PSD, calculating the average correlation of each time slice row with the new row of average PSD'"'"'s, and subtracting said average correlation from 1.

49. The apparatus of claim 35 wherein said means (a) through (e) are further defined as being capable of performing their indicated functions on a plurality of separate ECG channels.

50. The apparatus of claim 35 wherein said means (a) through (e) are further defined as being capable of performing their indicated functions on a plurality of ECG channels comprising at least a first, X, channel, a second Y, channel, and a third, Z, channel.

51. The apparatus of claim 50 further including means for calculating the mean value of the time-domain amplitude of each time slice to each of said plurality of ECG channels, prior to performing said frequency analysis.

52. The apparatus of claim 51 further including means for subtracting said mean time-domain value of each time slice from each data point value within that slice, thereby removing any D.C. offset in said time slice.

53. The apparatus of claim 52 further including means for summing the mean PSD values of each time slice row of each of said PSD matrices to form a fourth, average PSD matrix table.

54. The apparatus of claim 34 further including means for displaying tables of numbers derived from numbers in said matrix tables.

55. The apparatus of claim 54 further including means for displaying in three-dimensional-like contour plots the amplitude of said spectral components versus time and frequency.

56. The apparatus of claim 55 further including controller means for controlling in an automatic sequence the operation of each of said other means of said apparatus means in performing a frequency domain analysis of said ECG signals according to predetermined default parameters, said controller means being so structured as to facilitate manual change of the values of said parameters by an operator for performing a specific analysis for a specific purpose.

57. The apparatus of claim 56 wherein said selected parameters include:

(a) region of interest;

(b) pre-processing function;

(c) slice duration;

(d) sub-slice (step) duration;

(e) performing of mean subtraction;

(f) type of window function;

(g) number of FFT points; and

(h) orientation and magnification of threedimensional contour maps.

58. The apparatus of claim 55 wherein said means for displaying said spectral components includes visually discernible determination means to provide for correct patient risk classification.

59. The apparatus of claim 58 wherein said visually discernible determination means consists of:

visually discernible means for determining false positive patients wherein said spectral components, displayed by said means for displaying have smooth contours, and

visually discernible means for determining true positive patients wherein said spectral components, displayed by said means for displaying, have turbulent and disorganized contours.