HEART BEAT SIGNAL RECOGNITION

• US 20090043216A1
• Filed: 08/10/2007
• Published: 02/12/2009
• Est. Priority Date: 08/10/2007
• Status: Active Grant

• Current Assignee: Salutron, Inc.
• Original Assignee: Salutron, Inc.
• Inventors: Lin, Szming, Lo, Thomas Ying-Ching
• Granted Patent: US 8,046,058 B2
• Time in Patent Office: Days
• US Class Current: 600/501
• CPC Class Codes: A61B 5/0205 Simultaneously evaluating both cardiovascular conditions and different types of body conditions, e.g. heart and respiratory condition
• CPC Class Codes: A61B 5/35 by template matching
• CPC Class Codes: A61B 5/486 Bio-feedback A61B5/375 takes precedence
• CPC Class Codes: A61B 5/7214 using signal cancellation, e.g. based on input of two identical physiological sensors spaced apart, or based on two signals derived from the same sensor, for different optical wavelengths
• CPC Class Codes: A61B 5/7264 Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems neural networks per se G06N3/00; expert systems per se G06N5/00
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G06F 2218/08 Feature extraction
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

A subject'"'"'s heart rate is determined by recognizing heart beat patterns in a heart beat signal. A heart rate monitor receives a Doppler signal reflected from an artery of a target, performs demodulation and heart beat recognition techniques on the received signal to determine a set or sequence of features in each frame of the signal. Once a feature sequence is extracted from the signal, pattern classification is performed to determine if the extracted feature sequence is associated with one or more heart beats. The pattern classification may include finding the optimal state sequence by calculating the probability of each allowable state sequence based on the extracted feature sequence and heart beat models or additional noise models. Another pattern classification technique may determine a heart beat candidate using frame energy and dynamic thresholding methods followed by computing the probabilities between the feature sequence and each stored heart beat model or additional noise models. A further pattern classification technique may identify heart beat candidates using frame energy and dynamic thresholding methods and compute the similarity between the feature sequences and each of the stored heart beat templates. Post-processing is applied to heart beat candidates to determine if the candidates are associated with a true heartbeat, noise or some other signal source. Once a true heart beat is identified, the subject heart rate is updated based on the detected heart beat and displayed for a user.

Claims

1. A computer implemented method, comprising:

detecting a heart beat signal from a subject;

extracting two or more types of heart beat features from the heart beat signal;

determining a most likely state sequence for the heart beat signal based on the extracted features and the heart beat models;

identifying a heart beat based on the most likely state sequence; and

providing a heart rate for the subject based on said step of identifying the heart beat.

2. The computer implemented method of claim 1, wherein the heart beat signal is detected within an ultrasound signal reflected from the subject.

3. The computer implemented method of claim 1, wherein said step of extracting includes:

dividing the heart beat signal into one or more frames; and

extracting a set of features in each frame of the heart beat signal.

4. The computer implemented method if claim 1, wherein said step of extracting includes:

combining multiple types of features in a set of features.

5. The computer implemented method of claim 1, wherein said step of extracting includes:

performing linear predictive coding analysis on the heart beat signal to determine a set of linear predictive coding coefficients.

6. The computer implemented method of claim 1, wherein said step of extracting includes:

performing cepstral analysis using the set of linear predictive coding coefficients to generate a set of cepstral coefficients.

7. The computer implemented method of claim 6, wherein said step of extracting includes extracting Mel Frequency Cepstral Coefficients.

8. The computer implemented method of claim 6, wherein said step of extracting includes:

estimating one or more formants of the heart beat signal based on the set of the linear predictive coding coefficients.

9. The computer implemented method of claim 6, further comprising:

weighting a subset of the multiple types of features.

10. The computer implemented method of claim 8, wherein the multiple types of heart beat features include cepstral coefficients further include:

weighting the cepstral coefficients.

11. The computer implemented method of claim 1, further comprising:

calculating the delta features as a feature difference.

12. The computer implemented method of claim 1, further comprising:

accessing one or more heart beat models;

accessing one or more noise models; and

determining the most probable state sequence associated with the extracted feature sequence based on the models and grammar network using HMM techniques.

13. The computer implemented method of claim 1, wherein each of the noise models represent noise generated from muscle movement or body movement.

14. The computer implemented method of claim 1, wherein said step of updating a heart rate for the subject includes:

determining the most probable state sequence associated with the extracted feature sequence;

determining heart beat candidate by the most probable state sequence; and

updating the heart rate.

15. The computer implemented method of claim 1, wherein said step of updating a heart rate for the subject includes:

selecting a highest probability match between one or more probabilities generated for one or more heart beat models and noise models;

determining whether the highest probability is associated with one of the heart beat models; and

updating the heart rate based on the highest probability.

16. The computer implemented method of claim 14, wherein said step of updating a heart rate for the subject includes:

selecting the highest similarity match between the similarities generated for the heart beat templates; and

updating the heart rate if the highest similarity is higher than a selected threshold.

17. The computer implemented method of claim 14, further comprising:

Post-processing the heart beat candidate to eliminate an unlikely heart beat.

18. The computer implemented method of claim 15, wherein each of the heart beat models has two or more symbols.

19. The computer implemented method of claim 18, wherein the first symbol of a two-symbol heart beat model is derived from a first heart beat pattern associated with a heart contraction and the second symbol of a two-symbol heart beat model is derived from a second heart beat pattern associated with vessel backflow.

20. The computer implemented method of claim 15, wherein the heart beat models are derived from subjects having a different age.

21. The computer implemented method of claim 15, wherein heart beat models are derived from subjects having a different gender.

22. The computer implemented method of claim 15, wherein the heart beat models are derived from subjects having cardiac problems.

23. The computer implemented method of claim 18, wherein each symbol of a heart beat model has five states.

24. One or more processor readable storage devices having processor readable code embodied on said processor readable storage devices, said processor readable code for programming one or more processors to perform a method comprising:

extracting a set of features in every frame of a heart beat signal;

performing pattern classification; and

determining whether the feature sequence represents a true heart beat

25. The one or more processor readable storage devices of claim 24, wherein said step of extracting includes:

receiving a signal reflected from a subject'"'"'s artery;

dividing the signal into frames; and

extracting a set of heart beat features for the portion of the signal within each frame.

26. The one or more processor readable storage devices of claim 24, wherein the set of heart beat features may include LPC, cepstrum, MFCC, PLP, formant, energy, signal duration and delta features.

27. The one or more processor readable storage devices of claim 24, wherein said step of pattern classification includes:

using a Viterbi search algorithm to find the most likely state sequence of the extracted feature sequence.

28. The one or more processor readable storage devices of claim 24, wherein said step of pattern classification includes:

finding the heart beat candidate based on the energy and dynamic thresholding;

processing each value in the observation sequence against each state in each heart beat model and noise model; and

generating a total probability of match for each model by multiplying the probability of each value together for each model.

29. The one or more processor readable storage devices of claim 24, wherein said step of pattern classification includes:

finding the heart beat candidate based on the energy and dynamic thresholding; and

generating the similarity measure of the extracted feature sequence against each of the heart beat templates using dynamic time warping algorithm.

30. The one or more processor readable storage devices of claim 24, wherein said step of determining includes:

determining the subject'"'"'s heart rate based on a matched model, selected probability or selected similarity.

31. The one or more processor readable storage devices of claim 29, wherein the highest probability is associated with a heart beat model, said step of determining including:

determining if the heart beat candidate is associated with a true heart beat; and

updating the subject'"'"'s heart rate based on the heart beat signal.

32. The one or more processor readable storage devices of claim 29, wherein the highest probability is associated with a noise model, said step of determining including:

determining the feature sequence is associated with noise; and

ignoring the heart beat signal.

33. The one or more processor readable storage devices of claim 30, wherein said step of determining includes:

selecting the highest overall probability of match; and

determining if the selected overall probability of match meets a threshold value.

34. A heart rate monitor, comprising:

a transmitting element that transmits a signal towards a subject'"'"'s artery;

a receiving element which receives a reflected signal from the subject'"'"'s artery;

a feature extraction module which extracts features of the reflected signal; and

a pattern classification module able to classify the extracted feature sequence to determine whether the reflected signal is recognized as a heart beat.

35. The heart rate monitor of claim 34, wherein the transmitting element and receiving element are configured to operate at ultrasound frequencies.

36. The heart rate monitor of claim 34, a heart beat model store in communication with the pattern classification module, the pattern classification module configured to:

determine a most likely state sequence, probability of match or similarity of extracted feature sequence associated with the extracted feature sequence.

37. The heart rate monitor of claim 34, wherein one or more of the heart beat models have two symbols, each of the two symbol having five states which are used to classify the state sequence.

38. The heart rate monitor of claim 34, the feature extraction module configured to determine the energy, duration, formant and cepstral coefficients of a frame of the reflected signal.

39. The heart rate monitor of claim 34, the pattern classification module able to perform a hidden markov model analysis to determine the probability of match between the state sequence and one or more heart beat models and noise models.

40. The heart rate monitor of claim 34, further comprising:

a display for providing heart rate data, wherein the heart rate data is not changed if the feature sequence is not classified in a manner that recognizes the reflected signal as a heart beat.

41. A computer implemented method, comprising:

receiving a heart beat signal from a subject;

extracting features from a portion of the received signal;

performing pattern classification using the extracted features;

determining whether the heart beat signal is associated with a true heartbeat based on the pattern classification; and

updating a heart rate for the subject based on said step of determining the heart beat.

42. The computer implemented method of claim 41, wherein said step of performing pattern classification includes:

comparing the extracted features to one or more feature templates.

43. The computer implemented method of claim 42, wherein said step of comparing the extracted features includes:

selecting feature template associated with the smallest difference with the extracted features;

44. The computer implemented method of claim 41, wherein said step of performing pattern classification includes:

providing an observation sequence to one or more HMM heart beat models and noise models, the observation sequence including the extracted features.

45. The computer implemented method of claim 44, wherein said step of providing an observation sequence includes:

selecting the highest probability of match generated by the one or more HMM heart beat models and noise models.

46. The computer implemented method of claim 45, wherein said step of providing an observation sequence includes:

identifying whether the heart beat signal is a heart beat candidate based on the selected highest probability match

47. The computer implemented method of claim 41, wherein said step of performing pattern classification includes:

determining the highest probability path through a grammar network using one or more heart beat models and noise models and the extracted features.

48. The computer implemented method of claim 47, wherein said step of performing pattern classification includes:

determining whether the heart beat signal is a heart beat based on the selected path type.

49. The computer implemented method of claim 41, further comprising:

performing post processing to determine if a heart beat signal is associated with a true heart beat.

50. The computer implemented method of claim 49, wherein said step of performing post processing includes:

determining if a heart beat candidate associated with the extracted features occurs within a time range threshold.