Physiological signal denoising

• US 8,632,465 B1
• Filed: 11/03/2010
• Issued: 01/21/2014
• Est. Priority Date: 11/03/2009
• Status: Active Grant

• Current Assignee: VivaQuant LLC
• Original Assignee: VivaQuant LLC
• Inventors: Brockway, Marina
• Primary Examiner(s): Gedeon, Brian T
• Application Number:
  US12/938,995
• Publication Number:
  US 20140005988A1
• Time in Patent Office: 1,175 Days
• Field of Search: 600/300, 600/508, 600/509, 600/522, 600/523
• US Class Current: 600/300
• CPC Class Codes: A61B 5/316 Modalities, i.e. specific diagnostic methods
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle
• CPC Class Codes: A61B 5/7253 characterised by using transforms
• CPC Class Codes: G06F 17/14 Fourier, Walsh or analogous domain transformations , e.g. Laplace, Hilbert, Karhunen-Loeve, transforms for correlation function computation G06F17/156; spectrum analysers G01R23/16
• CPC Class Codes: G06F 18/2134 based on separation criteria, e.g. independent component analysis
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06F 2218/04 Denoising
• CPC Class Codes: G06F 2218/10 by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
• CPC Class Codes: H03H 17/0248 Filters characterised by a particular frequency response or filtering method

Abstract

Physiological signals are denoised. In accordance with an example embodiment, a denoised physiological signal is generated from an input signal including a desired physiological signal and noise. The input signal is decomposed from a first domain into subcomponents in a second domain of higher dimension than the first domain. Target subcomponents of the input signal that are associated with the desired physiological signal are identified, based upon the spatial distribution of the subcomponents. A denoised physiological signal is constructed in the first domain from at least one of the identified target subcomponents.

Claims

1. A method for computing a denoised physiological signal from an input signal including a desired pseudoperiodic physiological signal and noise, the method comprising:

decomposing the input signal from a first domain into subcomponents of the input signal in a second domain that is different than the first domain;

identifying target subcomponents of the input signal that are associated with the desired physiological signal based upon the time-based distribution of the subcomponents; and

reconstructing a denoised physiological signal in the first domain from at least two of the identified target subcomponents.

2. The method of claim 1, wherein

a cycle of the input signal is partitioned into at least two time windows based upon a characteristic band of frequencies associated with the desired signal within each of the at least two time windows, and

identifying target subcomponents includes:

identifying the locations of the at least two time windows based upon the location of a feature point of the cycle, and

identifying a target subcomponent in a time window as a subcomponent that contains more energy within the characteristic band of frequencies of the desired signal, than energy outside the characteristic band of frequencies.

3. The method of claim 1, wherein identifying target subcomponents of the input signal that are associated with the desired physiological signal based upon the time-based distribution of the subcomponents includes identifying target subcomponents using principal component analysis.

4. The method of claim 3, wherein identifying target subcomponents using principal component analysis includes

using principal component analysis to decorrelate the subcomponents, and

after decorrelating the subcomponents, identifying ones of the subcomponents as the target subcomponents.

5. The method of claim 1, wherein identifying target subcomponents of the input signal that are associated with the desired physiological signal includes

identifying noise subcomponents of the input signal as subcomponents having more noise energy than desired physiological signal energy, and

identifying, as target subcomponents, the subcomponents that are not noise subcomponents.

6. The method of claim 1, wherein identifying target subcomponents of the input signal that are associated with the desired physiological signal includes

identifying as noise subcomponents of the input signal, the subcomponents having more noise energy than desired physiological signal energy,

removing at least one of the noise subcomponents, and

identifying, as target subcomponents, the subcomponents remaining after removing the noise subcomponents.

7. The method of claim 1, wherein identifying target subcomponents of the input signal that are associated with the desired physiological signal includes identifying subcomponents having more desired physiological signal energy than noise energy.

8. The method of claim 1, wherein

the input signal has identifiable cycles that can be partitioned into time windows, each time window having an associated band of frequencies, and

identifying target subcomponents of the input signal that are associated with the desired physiological signal includes

identifying a cycle of the input signal,

partitioning the cycle into at least two time windows,

identifying noise subcomponents of the input signal in a time window as subcomponents that are not associated with a frequency band of a desired physiological signal within the time window, and

identifying the target subcomponents as subcomponents in the time window that are not identified as noise subcomponents.

9. The method of claim 1, wherein

the desired physiological signal is a mammalian ECG signal, and

identifying target subcomponents of the input signal that are associated with the desired physiological signal includes

identifying the location of a QRS complex in a cardiac cycle of the ECG,

identifying a first time window that includes the QRS complex,

identifying at least one time window in the cycle that does not include the QRS complex, and

identifying the target subcomponents, within the identified at least one time window, as subcomponents of the input signal that are associated with a frequency band of the desired physiological signal within the identified at least one window.

10. The method of claim 9, wherein identifying target subcomponents as subcomponents of the input signal that are associated with a frequency band of the desired physiological signal includes identifying subcomponents of the input signal having a spectral energy that substantially overlaps the frequency band.

11. The method of claim 1, wherein

the desired physiological signal is a mammalian ECG signal, and

identifying target subcomponents of the input signal that are associated with the desired physiological signal includes

identifying the location of a QRS complex in a cardiac cycle of the ECG,

identifying a first time window that includes the QRS complex,

identifying at least one time window in the cycle that does not include the QRS complex,

identifying as noise subcomponents, within the identified at least one time window, subcomponents of the input signal that are not associated with a frequency band of the desired physiological signal within the identified at least one window, and

identifying the target subcomponents as subcomponents in the identified at least one time window that were not identified as noise subcomponents.

12. The method of claim 1, wherein

the desired physiological signal is a mammalian ECG signal, and

identifying target subcomponents of the input signal that are associated with the desired physiological signal includes

identifying the location of a QRS complex in a cardiac cycle of the ECG,

identifying a first time window that includes the QRS complex,

identifying at least one time window in the cycle that does not include the QRS complex, and

comparing the frequency content of the subcomponents, within the at least one time window, with a characteristic band of frequencies identified via spectral analysis of a database containing recordings of representative ECG signals for the mammalian species, and identifying target subcomponents as subcomponents that have more power within the characteristic band of frequencies than power outside of the characteristic band of frequencies.

13. The method of claim 1, wherein decomposing the input signal from a first domain to a second domain, includes using one of:

a discrete cosine transform;

a wavelet related transform;

a Karhunen-Loeve transform;

a Fourier transform;

a Gabor transform; and

a filter bank.

14. The method of claim 1, wherein identifying target subcomponents includes using at least one of:

spatially selective filtering;

independent component analysis; and

periodic component analysis.

15. The method of claim 1, wherein

decomposing the input signal from a first domain into subcomponents in a second domain includes using a mathematical transform to transform the input signal from the first domain into the subcomponents, and

reconstructing a denoised physiological signal in the first domain includes combining at least two of the identified target subcomponents using an inverse of the mathematical transform.

16. The method of claim 1, wherein the input signal includes at least one of an electrocardiogram signal, a heart sound signal, a blood pressure signal, a respiratory information signal, a blood flow signal, a photoplethysmography signal, and a periodic stimulation response signal.

17. The method of claim 1, further including, before decomposing the input signal, segmenting the input signal according to the timing of an applied stimulation to assemble a plurality of signals.

18. The method of claim 1, wherein decomposing the input signal from a first domain into subcomponents of the input signal in a second domain includes decomposing the input signal from a first domain having a dimension size defined by a number of observed signal channels in the input signal, into a second domain having a dimension size defined by said number of channels multiplied by a number of subcomponents in each channel.

19. The method of claim 1,

further including partitioning a cycle of the input signal into at least two time windows based upon a characteristic band of frequencies of the desired signal within at least one of the time windows, at least one of said time windows containing a signal wave with wider-band frequency content than at least one other time window, and

wherein identifying the target subcomponents includes:

combining at least two subcomponents as a time-based function,

using the time-based function to identify a location of a signal wave with wider-band frequency content,

identifying the locations of the at least two time windows based upon the location of said signal wave, and

identifying a target subcomponent in a time window as a subcomponent that contains more energy within the characteristic band of frequencies of the desired signal within the window than energy outside the characteristic band of frequencies.

20. The method of claim 1, wherein identifying the target subcomponents includes partitioning a cycle of the input signal into at least two different time windows and selecting target subcomponents in each time window based upon a time-frequency representation of a cycle of the desired signal.

21. The method of claim 20, wherein the time-frequency representation of a cycle of the desired signal includes a spectral content distribution that varies synchronously with a feature point of the cycle.

22. The method of claim 20 wherein

the desired signal in each of the at least two time windows is associated with a characteristic band of frequencies, and

the target subcomponent is a subcomponent containing more energy inside the characteristic band of frequencies than energy outside the characteristic band of frequencies.

23. A method for computing a denoised ECG signal from an input signal including a desired ECG signal and noise, the method comprising:

decomposing the input signal into subcomponents;

identifying a location of the QRS complex of a cardiac cycle in the ECG signal;

identifying a first time window in the cardiac cycle that includes the QRS complex;

identifying at least one time window in the cardiac cycle that does not include the QRS complex;

for each of the identified time windows, identifying target subcomponents as subcomponents that contain more energy that is within the band of frequencies characteristic of the desired ECG signal in the time window than energy that is outside the band of frequencies of the desired ECG signal; and

reconstructing a denoised physiological signal using at least two of the identified target subcomponents.

24. The method of claim 23, wherein decomposing the input signal into subcomponents includes transforming the input signal from a first domain into subcomponents in a second domain using at least one of

a discrete cosine transform;

a wavelet related transform;

a Karhunen-Loeve transform;

a Fourier transform;

a Gabor transform; and

a filter bank.

25. The method of claim 23, wherein identifying target subcomponents includes identifying target subcomponents using at least one of:

principal component analysis;

independent component analysis;

spatially selective filtering; and

periodic component analysis.

26. The method of claim 23, wherein the desired ECG signal is an ECG of a species and identifying the band of frequencies characteristic of the desired ECG signal in each of the identified time windows includes:

identifying a database containing recordings of representative ECG signals for the species,

identifying the location of the QRS complex in cardiac cycles of the ECG,

partitioning the cardiac cycles into said identified time windows, and

using spectral analysis to identify the band of frequencies characteristic of the desired ECG signal in the identified time windows.

27. A system for denoising at least one of a P-wave and T-wave of an input signal including a desired ECG signal and noise, the system comprising:

a logic circuit; and

a memory circuit that stores instructions that, when executed by the logic circuit, carry out the steps of

decomposing the input signal into subcomponents,

identifying a location of the QRS complex of a cardiac cycle in the ECG signal,

identifying at least one time window in the cardiac cycle that includes the at least one of the P-wave and T-wave and does not include the QRS complex,

identifying target subcomponents in the at least one time window as subcomponents that contain more energy within the characteristic band of frequencies of the desired ECG signal within the at least one time window than energy outside the band of frequencies, and

reconstructing at least one of denoised P- and T-waves using the identified target subcomponents.

28. A method for removing noise from a non-QRS portion of a cardiac cycle of an input signal including a desired ECG signal and noise, the method comprising:

decomposing the input signal into subcomponents;

identifying a location of the QRS complex of a cardiac cycle in the ECG signal;

identifying at least one time window in the cardiac cycle that does not include the QRS complex;

identifying target subcomponents in the at least one time window based upon spectral energy of each target subcomponent relative to a characteristic spectral energy of the desired ECG signal within the at least one time window; and

reconstructing the non-QRS portion of the cardiac cycle using the identified target subcomponents in the at least one time window.

29. The method of claim 28 wherein the at least one time window spans the duration of the cardiac cycle outside the QRS complex.

30. The method of claim 28 wherein the input signal is from a species and identifying the target subcomponents includes:

identifying a characteristic spectral energy of the desired ECG signal in the at least one time window by characterizing the spectral energy of at least one of P-waves and T-waves of the species, and

identifying target subcomponents as subcomponents with spectral energy that overlaps said characteristic spectral energy.

31. The method of claim 30 wherein identifying target subcomponents as subcomponents with spectral energy that overlaps said characteristic spectral energy includes identifying subcomponents that, when combined, result in a quality of signal reconstruction greater than about 90%.

32. The method of claim 30 wherein identifying target subcomponents as subcomponents with spectral energy that overlaps said characteristic spectral energy includes identifying subcomponents that, when combined, result in a quality of signal reconstruction greater than about 95%.