Analyzing electrocardiogram signals with multiscale short-time fourier transforms

• US 8,494,622 B2
• Filed: 03/23/2012
• Issued: 07/23/2013
• Est. Priority Date: 02/03/2010
• Status: Active Grant

• Current Assignee: National Instruments Corporation (Emerson Electric Company)
• Original Assignee: National Instruments Corporation (Emerson Electric Company)
• Inventors: Gu, Zhijun, Zhang, Nanxiong, Qian, Shie
• Primary Examiner(s): Schaetzle, Kennedy
• Application Number:
  US13/428,047
• Publication Number:
  US 20120179059A1
• Time in Patent Office: 487 Days
• Field of Search: None
• US Class Current: 600/509
• CPC Class Codes: A61B 5/316 Modalities, i.e. specific diagnostic methods
• CPC Class Codes: A61B 5/339 Displays specially adapted therefor
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle
• CPC Class Codes: A61B 5/352 Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval
• CPC Class Codes: A61B 5/366 Detecting abnormal QRS complex, e.g. widening
• CPC Class Codes: A61B 5/7253 characterised by using transforms
• CPC Class Codes: A61B 5/7257 using Fourier transforms

Abstract

A system, method, and memory medium for operating on an electrocardiogram (ECG) signal. A multiscale short-time Fourier transform (STFT) is perform on a set of ECG samples {s(n)} to obtain a transform array. For each sufficiently energetic peak in the transform array, a refined window width value and a refined window displacement value is generated by: computing an inner product between the set of samples and each of a plurality of functions, where the plurality of functions are sufficiently close to a coarse approximation function given by the peak location; and solving a linear system Av=c for the unknown vector v, where the vector c is determined by the inner products, where the matrix A is determined by the center times of the plurality of functions. After appropriate selection, the refined window width and refined window displacement may be used to represent ECG waveform features.

Claims

1. A non-transitory computer-readable memory medium that stores program instructions executable by a computer system to:

receive a discrete electrocardiogram (ECG) signal, wherein the discrete ECG signal includes one or more waveform features indicative of cardiac condition;

apply a sliding window multiscale frequency transform to the discrete electrocardiogram (ECG) signal, thereby generating a transform array parameterized by a specified plural set of window features;

determine a plurality of peaks in the transform array, including determining coarse values for each of the specified plural set of window features for each peak;

determine, for each peak, at least three windowed functions, wherein each of the windowed functions has window features with values approximately equal to the coarse values corresponding to the specified plural set of window features; and

determine, for each peak, refined values for the plural set of window features based on the discrete ECG signal and the at least three windowed functions;

wherein the refined values for the plural set of window features are indicative of a specified cardiac condition.

2. The non-transitory computer-readable memory medium of claim 1, wherein the specified plural set of window features comprises window width, window displacement τ

and frequency ω

;

wherein to apply a sliding window multiscale frequency transform to the discrete electrocardiogram (ECG) signal, the program instructions are executable to compute a plurality of short-time Fourier transforms (STFT), wherein each of the STFTs is computed using a different value of window width, and wherein each of the STFTs is a function of window displacement time τ and

frequency ω

.

3. The non-transitory computer-readable memory medium of claim 2, wherein to determine refined values for the plural set of window features for each peak, the program instructions are further executable to:

compute, for each peak, an inner product of the discrete ECG signal and each of the at least three functions, wherein the at least three functions are windowed sinusoids (a) having a common frequency equal to the frequency value of the peak, (b) being centered at times that are in a neighborhood of the window displacement value of the peak, and (c) having a common width that is based on the window width value of the peak; and

solve a linear system Av=c for the unknown vector v, wherein the vector c is determined by the inner products, wherein the matrix A is determined by the times that are in the neighborhood of the window displacement value of the peak.

4. The non-transitory computer-readable memory medium of claim 3, wherein the windowed sinusoids are Gaussian-modulated complex exponentials.

5. The non-transitory computer-readable memory medium of claim 2, wherein to determine the plurality of peaks in the transform array, the program instructions are further executable to determine an amplitude value for each peak, and wherein the program instructions are further executable to:

identify a first of the peaks as corresponding to a QRS complex based on a determination that the amplitude value of the first peak exceeds a threshold value and that the refined window displacement value for the first peak satisfies one or more time constraints;

compute a QRS onset time and/or a QRS offset time based on the refined window displacement value and the refined window width of the first peak; and

display the QRS onset time and/or the QRS offset time via a display device.

6. The non-transitory computer-readable memory medium of claim 5, wherein the discrete ECG signal comprises a first set of samples of an ECG signal, wherein the program instructions are further executable to:

repeat said receiving, said applying, said determining a plurality of peaks, said determining at least three windowed functions, and said determining refined values for a series of further sets of samples of the ECG signal.

7. The non-transitory computer-readable memory medium of claim 5, wherein the threshold value is an adaptive threshold that changes over time.

8. The non-transitory computer-readable memory medium of claim 5, wherein the one or more time constraints include a time constraint on the refined window displacement value of the first peak relative to a time of a previously identified feature in the discrete ECG signal.

9. The non-transitory computer-readable memory medium of claim 2, wherein the program instructions are further executable to:

compute, for each peak P_j of the plurality of peaks, a windowed sinusoid function h_j that is determined by the refined window width value of the peak P_j, the refined window displacement value of the peak P_j, and the frequency value of the peak P_j;

generate a linear combination of the windowed sinusoid functions h_j; and

display the real part of the linear combination via a display device.

10. The non-transitory computer-readable memory medium of claim 9, wherein the windowed sinusoids are Gaussian-modulated complex exponentials.

11. The non-transitory computer-readable memory medium of claim 1, wherein the discrete ECG signal is a human ECG signal, wherein the one or more waveform features include one or more of:

a P wave, a QRS complex, or a T wave.

12. The non-transitory computer-readable memory medium of claim 11, wherein the program instructions are further executable to:

display the waveform feature or a visual indication of the waveform feature via a display device.

13. The non-transitory computer-readable memory medium of claim 1, wherein to determine a plurality of peaks in the transform array the program instructions are further executable to search for N_P peaks of largest absolute value in the transform array, wherein N_P is a positive integer.

14. The non-transitory computer-readable memory medium of claim 13, wherein said searching is restricted to a band of frequencies that is smaller than the frequency range of the sliding window multiscale frequency transform.

15. A non-transitory computer-readable memory medium storing program instructions executable by a computer system to:

receive a discrete electrocardiogram (ECG) signal, wherein the discrete ECG signal includes a waveform feature indicative of cardiac condition;

compute a plurality of windowed multiscale frequency transforms based on the discrete ECG signal, wherein each windowed multiscale frequency transform is computed using a respective value of window width;

determine an initial set of parameter values that characterize the waveform feature based on the plurality of windowed multiscale frequency transforms;

determine at least three windowed functions, wherein each of the windowed functions has window features with values approximately equal to the initial set of parameter values;

compute at least two refined parameter values of the waveform feature based on the at least three windowed functions and the initial set of parameter values; and

store the at least two refined parameter values of the waveform feature in a memory.

16. The non-transitory computer-readable memory medium of claim 15,

wherein the initial set of parameter values includes a first window displacement time τ

₀, a first frequency ω

₀ and a first window width σ

₀; and

wherein the plurality of windowed multiscale frequency transforms comprise a plurality of short-time Fourier transforms (STFT), wherein each of the STFTs is computed using a different value of window width, and wherein each of the STFTs is a function of window displacement time τ and

frequency ω

.

17. The non-transitory computer-readable memory medium of claim 15, wherein the plurality of STFTs are computed using Gaussian windows.

18. The non-transitory computer-readable memory medium of claim 15, wherein to determine an initial set of parameter values that characterize the waveform feature, the program instructions are executable to search the STFTs for a local maximum or global maximum of absolute value in the STFTs.

19. The non-transitory computer-readable memory medium of claim 18, wherein the discrete ECG signal is a human ECG signal, wherein said searching is restricted to frequencies in the interval from 10 Hz to 25 Hz.

20. The non-transitory computer-readable memory medium of claim 15,

wherein the at least three functions are windowed sinusoids (a) having a common frequency equal to the first frequency ω

₀, (b) being centered at times that are in a neighborhood of the first window displacement time τ

₀, and (c) having a common width that is based on the first window width σ

₀;

wherein to compute the at least two refined parameter values of the waveform feature based on the at least three windowed functions and the initial set of parameter values, the program instructions are executable to;

compute an inner product of the discrete ECG signal with each of the at least three functions to obtain respectively a first complex value, a second complex value and a third complex value; and

compute a time of the waveform feature and a duration of the waveform feature based on the first, second and third complex values and based on said times in the neighborhood of the first window displacement time τ

₀.

21. The non-transitory computer-readable memory medium of claim 16, wherein the waveform feature is a QRS complex, wherein the program instructions are further executable to:

compute an onset time and/or an offset time of the QRS complex based on the waveform feature time and the waveform feature duration.

22. The non-transitory computer-readable memory medium of claim 20, wherein the waveform feature is a QRS complex, a T wave or a P wave.

23. The non-transitory computer-readable memory medium of claim 16, wherein the windowed sinusoids are Gaussian-modulated complex exponentials.

24. The non-transitory computer-readable memory medium of claim 20, wherein said computing the waveform feature time and the waveform feature duration includes solving a linear system Av=c for the unknown vector v, wherein the matrix A is determined by the times in the neighborhood of the first window displacement time τ

₀, wherein the vector c is determined by the first, second and third complex values.

25. The non-transitory computer-readable memory medium of claim 24, wherein the vector c is determined by the expression:

c=[ln|z₀|/|z<sub>−

1</sub>|,ln|z₀|/|z₊₁|]^T,

wherein z₀, z<sub>−

1</sub> and z₊₁ are the first, second and third complex values, respectively.

26. The non-transitory computer-readable memory medium of claim 20, wherein the times that are in the neighborhood of the first window displacement time τ

₀ include τ

₀, τ

₀+Δ

τ and

τ

₀−

Δ

τ

, where Δ

τ

is a positive real number.

27. A computer-implemented method comprising:

utilizing a computer system to perform;

receiving a discrete electrocardiogram (ECG) signal, wherein the discrete ECG signal includes a waveform feature indicative of cardiac condition;

computing a plurality of windowed multiscale frequency transforms based on the discrete ECG signal, wherein each windowed multiscale frequency transform is computed using a respective value of window width;

determining an initial set of parameter values that characterize the waveform feature based on the plurality of windowed multiscale frequency transforms;

determining at least three windowed functions, wherein each of the windowed functions has window features with values approximately equal to the initial set of parameter values;

computing at least two refined parameter values of the waveform feature based on the at least three windowed functions and the initial set of parameter values; and

storing the at least two refined parameter values of the waveform feature in a memory.