Method and apparatus for low complexity spectral analysis of bio-signals

Patent metadata

Field	Value
Publication number	US-9760536-B2
Application number	US-201314421719-A
Country	US
Kind code	B2
Filing date	Aug 15, 2013
Priority date	Aug 16, 2012
Publication date	Sep 12, 2017
Grant date	Sep 12, 2017

Abstract

A method and device for reducing the computational complexity of a processing algorithm, of a discrete signal, in particular of the spectral estimation and analysis of bio-signals, with minimum or no quality loss, which comprises steps of (a) choosing a domain, such that transforming the signal to the chosen domain results to an approximately sparse representation, wherein at least part of the output data vector has zero or low magnitude elements; (b) converting the original signal in the domain chosen in step (a) through a mathematical transform consisting of arithmetic operations resulting in a vector of output data; (c) reformulating the processing algorithm of the original signal in the original domain into a modified algorithm consisting of equivalent arithmetic operations in the domain chosen in step (a) to yield the expected result with the expected quality quantified in terms of a suitable application metric; (d) combining the mathematical transform of step (b) and the equivalent mathematical operations introduced in step (c) for obtaining the expected result within the original domain with the expected quality; (e) selecting a threshold value based on the difference in the mean magnitude value of the elements of the output data vector of the transform said in step (b) and the preferred complexity reduction and degree of output quality loss that can be tolerated in the expected result within the target application; (f) pruning a number of elements the magnitude of which is less than the threshold value selected in step (e); and/or eliminating arithmetic operations associated with the pruned elements of step (f) either in the mathematical transform of step (b) and/or in the equivalent algorithm of step (c).

First claim

The invention claimed is:

1. A method for estimating a frequency spectrum of a bio-signal that is non-uniformly spaced with low complexity, the method comprising the steps of:
(a) sampling the bio-signal with a predefined sized window function;
(b) replacing a sampled data value, within the window, at an arbitrary point by several data values on a regular mesh in such a way that sums over the mesh are an accurate approximation to sums over the original arbitrary point;
(c) transforming the signal consisting of the extrapolated data samples of step (b) within the window in the wavelet domain using a wavelet transform by decomposing the signal into multiple frequency resolutions representing different classes of data by applying a series of suitable high-pass and low-pass filters based on the used wavelet basis;
(d) processing the output data vector of the applied wavelet transform using reformulated Fourier operations with modified twiddle factors, values of the modified twiddle factors are obtained from the frequency response of the high-pass and low-pass filters of the applied wavelet transform;
(e) selecting a threshold value such that the data obtained from the high-pass filters of the applied wavelet transform are pruned after applying the selected threshold;
(f) eliminating the operations associated with the pruned high-pass data of step (e), in part or as a whole;
(g) evaluating sine and cosine functions at the times corresponding to the data samples within the taken window using the output data obtained after applying steps (c), (d) and (e);
(h) computing a time shift for each frequency in a desired set of frequencies to orthogonalize the sine and cosine components; and
(i) computing the power of each frequency in a desired set of frequencies based on the estimated amplitudes of the evaluated sine and cosine functions.

2. The method of claim 1 , wherein
the biosignal includes time intervals of consecutive heart beats;
step (b) includes an extrapolation method using a Fast-Lomb method;
in step (c) a Haar Wavelet Transform with one frequency resolution level is used to transform the sampled data within the wavelet transform;
in step (d) the output data vector of a one-stage Haar Wavelet transform is processed by reformulated Fourier operations with modified twiddle factors, the values of which are obtained from the frequency response of the high-pass and low-pass filters used in the Haar wavelet transform;
in step (e) the data obtained from the high-pass filter of the Haar wavelet transform are pruned in part or as a whole;
in step (f) the operations associated with the pruned high-pass data of step (e) are eliminated, in part or as a whole;
in step (h) computing a time shift for each frequency in a desired set of frequencies to orthogonalize the sine and cosine components is equivalent to the time shift applied in a least-squares analysis method and/or a Fast-Lomb method; and
in step (i) computing the power of each frequency in a desired set of frequencies based on the estimated amplitudes of the sine and cosine functions is performed as in a least-squares analysis method.

3. The method of claim 2 , wherein
the modified twiddle factors of the reformulated Fourier operations that correspond to the frequency response of the low-pass filters used in the applied Haar wavelet transform are sorted based on their magnitude; and
the factors with smaller magnitude than the rest of the twiddle factors and the associated operations are being eliminated.

4. The method of claim 1 for estimating the time-frequency distribution of the biosignal that is non-uniformly spaced with low complexity, the method further comprising the steps of:
(j) recording a signal;
(k) splitting up the recorded signal into overlapping segments;
(l) windowing the overlapping segments;
(m) using a sliding window configuration in order to process the data in different time instances;
(n) applying steps (b) to (i) to compute the frequency spectrum for each segment; and
(m) normalizing each processed segment equally by time-averaging the individual spectrum; achieved by applying a normalizing factor to each processed segment.

5. The method of claim 1 for analyzing the frequency or time-frequency spectrum of the biosignal that is non-uniformly spaced within a set of desired frequencies with low complexity comprises the steps of:
(j) estimating the frequency spectrum or time-frequency according to steps (a) to (j);
(k) calculating the ratio between the low frequency power and high frequency power within a set of desired low and high frequencies; and
(l)comparing the estimated ratio with a threshold value for evaluating various conditions of interest.

6. The method of claim 5 , wherein
in step (j), the methods are applied for estimating the frequency spectrum;
in step (k) the ratio between the low frequency power and high frequency power within a set of desired low and high frequencies; the desired range being usually 0.04-0.15 Hz for the low frequencies and 0.15-0.4 Hz for the high frequencies;
the differences between the estimated ratio without any pruning and the ratio obtained after applying any pruning calculated and used as a quality metric; and
in step (l) the ratio is compared with a proper threshold value in order to indicate any cardiac malfunction.

7. The method of claim 1 , further comprising the steps of:
(j) selecting threshold values based on differences in the magnitude of the various classes of data in the wavelet domain;
(k) selecting threshold values based on the magnitudes of the twiddle factors that correspond to the frequency response of the filters used wavelet transform;
(l) comparing the resulted outputs of the wavelet transform with the determined thresholds of step (a) and dropping only those that are less than the determined threshold; and
(m) comparing the computed value in the modified Fourier transform with the determined thresholds of step (b) and dropping only those that are less than the determined threshold.

8. The method for combining the proposed power spectral estimation and analysis according to claim 1 with other wavelet based signal analysis and processing tools, the method comprising:
(j) sampling the input data within a window, the input data representing the biosignal;
(k) decomposing the recorded signals by a wavelet transform through the application of filtering operations into multiple frequency resolution levels representing different classes of data;
(l) using the outputs of a preferred frequency resolution level as input to the modified Fourier operations described in above claims;
(m) estimating the power spectra and/or time frequency distribution in according to steps (a) to (i); and
(n) using the output of the applied wavelet transform for further signal analysis and processing tool.

9. The method of claim 8 , wherein
in step (n) the proposed method is combined with the compression of heart rate data, or extraction of heart characteristics using for instance a wavelet based delineation algorithm;
the pruning of data or elimination of operation is applied.

10. A method for re-using the wavelet stages of other cardiac analysis and processing tools with the spectral analysis method of claim 8 comprising:
(o) evaluating various wavelet basis for delineation of cardiac data and power spectral analysis;
(p) selecting the wavelet basis that result in large complexity reduction and minor or no loss in output quality of the delineation and spectral analysis algorithms; and
(q) using the wavelet transform based on the selected basis for performing both delineation and spectral estimation and analysis of the preferred signal.

Assignees

• Ecole Polytechnique Fed De Lausanne (Epfl)
• Ecole Polytechnique Fed Lausanne Epfl

Inventors

• Karakonstantis Georgios
• Sankaranarayanan Aviinaash
• Burg Andreas
• Murali Srinivasan
• Atienza Alonso David

Classifications

• G06F17/142 Primary

  Fast Fourier transforms, e.g. using a Cooley-Tukey type algorithm · CPC title

• G06F17/148

  Wavelet transforms · CPC title

• A61B5/7257

  using Fourier transforms · CPC title

• A61B5/726

  using Wavelet transforms · CPC title

• A61B5/7253

  characterised by using transforms · CPC title