Motion and noise artifact detection and reconstruction algorithms for photoplethysmogram and equivalent signals

1. A computer implemented method for physiological parameter monitoring using a signal used as a Photoplethysmogram (PPG) signal, the computer implemented method comprising:

obtaining a time frequency spectrum of a segment of the signal used as the PPG signal;

obtaining, from the time frequency spectrum, a noise quality index for the segment; the noise quality index being used to determine whether the segment is corrupted by motion and noise artifacts;

wherein obtaining a noise quality index comprises:

determining a dominant frequency in the time frequency spectrum of the segment;

normalizing the time frequency spectrum to a total power in a narrow band centered at the dominant frequency;

determining a first trace of amplitudes in the narrow band spectrum of the time frequency spectrum centered at the dominant frequency;

determining a second trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at twice the dominant frequency;

determining a third trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at three times the dominant frequency;

subtracting the first, second and third traces of amplitudes from the time frequency spectrum;

obtaining, after subtracting, a total power remaining in the time frequency spectrum, said total power remaining referred to as a residual noise power;

determining a difference in frequency between the first trace and the second and third traces, the difference in frequency referred to as a projected difference;

the noise quality index being a weighted sum of factors including the residual noise power and the projected difference; weights being selected such that each weighted factor represents less than a predetermined percentage of power in an uncorrupted segment;

applying a statistical learning method, using the noise quality index, to determine whether the segment is corrupted by motion and noise artifacts or not corrupted by motion and noise artifacts;

and,

if motion and noise artifacts are not present, including the segment in determination of a physiological parameter.

2. The computer implemented method of claim 1 wherein the statistical learning method is a Support Vector Machine (SVM).

3. The computer implemented method of claim 1 further comprising determining a difference between the dominant frequency and a heart rate obtained from peak to peak intervals from the signal used as the PPG signal in a time domain, the difference between the dominant frequency and the heart rate obtained from peak to peak intervals referred to as a heart rate frequency difference; wherein the noise quality index also comprises the heart rate frequency difference; the noise quality index being a weighted sum of factors including the residual noise power, the projected difference and the heart rate frequency difference.

4. The computer implemented method of claim 3 wherein the statistical learning method is a Support Vector Machine (SVM).

5. The computer implemented method of claim 1 wherein the signal used as the PPG signal is preprocessed before obtaining the time frequency spectrum.

6. The computer implemented method of claim 5 wherein the signal used as the PPG signal is preprocessed by filtering with a bandpass filter of predetermined band width.

7. The computer implemented method of claim 1 wherein the time frequency spectrum is obtained using variable frequency complex demodulation.

8. A computer implemented method for physiological parameter monitoring using a signal used as a Photoplethysmogram (PPG) signal, the computer implemented method comprising:

obtaining a time frequency spectrum of a segment of the signal used as the PPG signal;

obtaining, from the time frequency spectrum, a noise quality index for the segment; the noise quality index being used to determine whether the segment is corrupted by motion and noise artifacts;

wherein obtaining a noise quality index comprises:

determining a dominant frequency in the time frequency spectrum of the segment;

normalizing the time frequency spectrum to a total power in a narrow band centered at the dominant frequency;

determining a first trace of amplitudes in the narrow band spectrum of the time frequency spectrum centered at the dominant frequency;

determining a second trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at twice the dominant frequency;

determining a third trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at three times the dominant frequency;

subtracting the first, second and third traces of amplitudes from the time frequency spectrum;

obtaining, after subtracting, a total power remaining in the time frequency spectrum, said total power remaining referred to as a residual noise power;

determining a difference between the dominant frequency and a heart rate obtained from peak to peak intervals from the signal used as the PPG signal in a time domain, the difference between the dominant frequency and the heart rate obtained from peak to peak intervals referred to as a heart rate frequency difference;

the noise quality index also comprising the heart rate frequency difference; the noise quality index being a weighted sum of factors including the residual noise power and the heart rate frequency difference; weights being selected such that each weighted factor represents less than a predetermined percentage of power in an uncorrupted segment;

applying a statistical learning method, using the noise quality index, to determine whether the segment is corrupted by motion and noise artifacts or not corrupted by motion and noise artifacts;

and,

if motion and noise artifacts are not present, including the segment in determination of a physiological parameter.

9. The computer implemented method of claim 8 wherein the statistical learning method is a Support Vector Machine (SVM).

10. A system for physiological parameter monitoring using a signal used as a Photoplethysmogram (PPG) signal, the system comprising:

a processor configured to:

receive the signal used as the PPG signal;

obtain a time frequency spectrum of a segment of the signal used as the PPG signal;

obtain, from the time frequency spectrum, a noise quality index for the segment; the noise quality index being used to determine whether the segment is corrupted by motion and noise artifacts;

determine a dominant frequency in the time frequency spectrum of the segment;

normalize the time frequency spectrum to the total power in a narrow band centered at the dominant frequency;

determine a first trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at the dominant frequency;

determine a second trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at twice the dominant frequency;

determine a third trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at three times the dominant frequency;

subtract the first, second and third traces of amplitudes from the time frequency spectrum;

obtain, after subtracting, a total power remaining in the time frequency spectrum; the total power remaining referred to as a residual noise power;

determine a difference in frequency between the first trace and the second and third traces; the difference in frequency referred to as a projected difference;

apply a statistical learning method, using the noise quality index, to determine whether the segment is corrupted by motion and noise artifacts or not corrupted by motion and noise artifacts;

and,

if motion and noise artifacts are not present, include the segment in determination of a physiological parameter;

the noise quality index being a weighted sum of factors including the residual noise power and the projected difference; weights being selected such that each weighted factor represents less than a predetermined percentage of power in an uncorrupted segment.

11. The system of claim 10 wherein the statistical learning method is a Support Vector Machine (SVM).

12. The system of claim 10 wherein the processor is further configured to determine a difference between the dominant frequency and a heart rate obtained from peak to peak intervals from the signal used as the PPG signal in a time domain; the difference between the dominant frequency and the heart rate obtained from peak to peak intervals referred to as a heart rate frequency difference; wherein the noise quality index also comprises the heart rate frequency difference; and, wherein the noise quality index is a weighted sum of factors including the residual noise power, the projected difference and the heart rate frequency difference.

13. The system of claim 12 wherein the statistical learning method is a Support Vector Machine (SVM).

14. The system of claim 10 wherein the processor is further configured to preprocess the signal used as the PPG signal before obtaining the time frequency spectrum.

15. The system of claim 14 wherein the processor is configured to preprocess the signal used as the PPG signal by filtering with a bandpass filter of predetermined bandwidth.

16. The system of claim 10 wherein the processor is configured to obtain the time frequency spectrum using variable frequency complex demodulation.

17. A system for detection of motion and noise artifacts in a signal used as a Photoplethysmogram (PPG) signal, the system comprising:

a processor configured to:

receive the signal used as the PPG signal;

obtain a time frequency spectrum of a segment of the signal used as the PPG signal;

obtain, from the time frequency spectrum, a noise quality index for the segment; the noise quality index being used to determine whether the segment is corrupted by motion and noise artifacts;

determine a dominant frequency in the time frequency spectrum of the segment;

normalize the time frequency spectrum to the total power in a narrow band centered at the dominant frequency;

determine a first trace of amplitudes in the narrow band spectrum of the time frequency spectrum centered at the dominant frequency;

determine a second trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at twice the dominant frequency;

determine a third trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at three times the dominant frequency;

subtract the first, second and third traces of amplitudes from the time frequency spectrum;

obtain, after subtracting, a total power remaining in the time frequency spectrum; the total power remaining referred to as a residual noise power;

determine a difference between the dominant frequency and a heart rate obtained from peak to peak intervals from the signal used as the PPG signal in a time domain; the difference between the dominant frequency and the heart rate obtained from peak to peak intervals referred to as a heart rate frequency difference;

the noise quality index being a weighted sum of factors including the residual noise power and the heart rate frequency difference; weights being selected such that each weighted factor represents less than a predetermined percentage of power in an uncorrupted segment;

apply a statistical learning method, using the noise quality index, to determine whether the segment is corrupted by motion and noise artifacts or not corrupted by motion and noise artifacts;

and,

if motion and noise artifacts are not present, include the segment in determination of a physiological parameter.

18. The system of claim 17 wherein the statistical learning method is a Support Vector Machine (SVM).

19. A computer program product comprising:

tangible computer usable media, having computer readable code embodied therein, the computer readable code, when executed by one or more processors, causes the one or more processors to:

obtain a time frequency spectrum of a segment of a signal used as a PPG signal;

obtain, from the time frequency spectrum, a noise quality index for the segment; the noise quality index being used to determine whether the segment is corrupted by motion and noise artifacts;

determine a dominant frequency in the time frequency spectrum of the segment;

normalize the time frequency spectrum to the total power in a narrow band centered at the dominant frequency;

determine a first trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at the dominant frequency;

determine a second trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at twice the dominant frequency;

determine a third trace of amplitudes in a narrow band spectrum of the time frequency spectrum centered at three times the dominant frequency;

subtract the first, second and third traces of amplitudes from the time frequency spectrum; and

obtain, after subtracting, a total power remaining in the time frequency spectrum; the total power remaining referred to as a residual noise power;

determine a difference in frequency between the first trace and the second and third traces; the difference in frequency referred to as a projected difference;

apply a statistical learning method, using the noise quality index, to determine whether the segment is corrupted by motion and noise artifacts or not corrupted by motion and noise artifacts;

and,

if motion and noise artifacts are not present, include the segment in determination of a physiological parameter;

wherein the noise quality index is a weighted sum of factors including the residual noise power and the projected difference; weights being selected such that each weighted factor represents less than a predetermined percentage of power in an uncorrupted segment.

20. The computer program product of claim 19 wherein the statistical learning method is a Support Vector Machine (SVM).