Machine learnt model to detect REM sleep periods using a spectral analysis of heart rate and motion

• US 10,470,719 B2
• Filed: 01/31/2017
• Issued: 11/12/2019
• Est. Priority Date: 02/01/2016
• Status: Active Grant

• Current Assignee: Verily Life Sciences LLC (Alphabet Inc.)
• Original Assignee: Verily Life Sciences LLC (Alphabet Inc.)
• Inventors: Shimol, David Ben, Shoeb, Ali
• Primary Examiner(s): Lee, Paul D
• Application Number:
  US15/421,349
• Publication Number:
  US 20170215808A1
• Time in Patent Office: 1,015 Days
• Field of Search: 702181
• CPC Class Codes: A61B 2562/0219 Inertial sensors, e.g. accelerometers, gyroscopes, tilt switches
• CPC Class Codes: A61B 5/0002 Remote monitoring of patients using telemetry, e.g. transmission of vital signals via a communication network A61B5/07 takes precedence; transmission systems for measured values G08C; transmission H04B; transmission of digital information H04L; wireless communication networks H04W
• 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 5/02405 Determining heart rate variability
• CPC Class Codes: A61B 5/02416 using photoplethysmograph signals, e.g. generated by infrared radiation A61B5/14552 takes precedence
• CPC Class Codes: A61B 5/11 Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb for measuring pulse A61B5/02 ;A61B5/1038 takes precedence; motion detection to correct for motion artifacts in physiological signals A61B5/721
• CPC Class Codes: A61B 5/4809 Sleep detection, i.e. determining whether a subject is asleep or not
• CPC Class Codes: A61B 5/4812 Detecting sleep stages or cycles
• CPC Class Codes: A61B 5/7246 using correlation, e.g. template matching or determination of similarity
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: A61B 5/7275 Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

The present disclosure relates to systems and methods for probabilistically estimating an individual'"'"'s sleep stage based on spectral analyses of pulse rate and motion data. In one implementation, the method may include receiving signals from sensors worn by the individual, the signals including a photoplethysmographic (PPG) signal and an accelerometer signal; dividing the PPG signal into segments; determining a beat interval associated with each segment; resampling the set of beat intervals to generate an interval signal; and generating signal features based on the interval signal and the accelerometer signal, including a spectrogram of the interval signal. The method may further include determining a sleep stage for the individual by comparing the signal features to a sleep stage classifier included in a learning library. The sleep stage classifier may include one or more functions defining a likelihood that the individual is in the sleep stage based on the signal features.

Claims

1. A method for probabilistically determining an individual'"'"'s sleep stage, the method comprising the following operations performed via one or more processors:

receiving a set of signals from a set of sensors worn by the individual, the set of signals including a photoplethysmographic (PPG) signal and an accelerometer signal;

dividing the PPG signal into a set of equally-timed segments if the received PPG signal comprises segments having unequal time durations;

determining a beat interval associated with each segment, the beat interval reflecting an elapsed time between successive heartbeats;

sampling the set of beat intervals to generate an interval signal;

generating a set of signal features based on the interval signal and the accelerometer signal, the set of signal features including a spectrogram of the interval signal; and

determining a sleep stage for the individual by operating on the set of signal features using a sleep stage classifier included in a learning library, wherein the sleep stage classifier comprises a set of functions defining a likelihood that the individual is in the sleep stage based on the set of signal features.

2. The method of claim 1, wherein sampling the set of beat intervals comprises sampling using the frequency of the equally-timed segments.

3. The method of claim 1, wherein sampling the set of beat intervals comprises resampling using a frequency different from the frequency of the equally-time segments.

4. The method of claim 1, wherein the set of functions includes signal weights, and wherein each signal weight is based on a correlation between a feature in the set of signal features and a known sleep stage.

5. The method of claim 4, further comprising:

sampling the accelerometer signal at a first sampling rate;

differentiating the sampled accelerometer signal over each channel associated with the accelerometer signal; and

averaging a magnitude of the differentiated accelerometer signals over an accelerometer time window to generate a time-averaged motion signal;

wherein the set of signal features further comprises the time-averaged motion signal such that the sleep stage is further based on the time-averaged motion signal.

6. The method of claim 5, wherein the learning library comprises data from among a plurality of sessions for a plurality of individuals, and wherein the set of signal weights is based on normalizing each feature in the set of features over each session for each of the plurality of individuals.

7. The method of claim 4, wherein the set of signal features further comprise an average pulse rate derived from the PPG signal sampled at a second sampling rate and a sum of the absolute energy in each of a plurality of frequency bands in the spectrogram of the interval signal.

8. The method of claim 7, wherein the sleep stage is determined from among a set of potential sleep stages, wherein the set of potential sleep stages comprises an awake stage, a non-REM stage, and a REM stage, and further comprising determining a sleep stage likelihood for each of the potential sleep stages based on the set of signal features and the set of signal weights in the sleep stage classifier.

9. The method of claim 8, wherein determining the sleep stage further comprises:

identifying a most likely sleep stage based on a highest sleep stage likelihood among a set of sleep stage likelihoods including each sleep stage likelihood associated with a potential sleep stage; and

determining whether the highest sleep stage likelihood exceeds a preset stage likelihood threshold.

10. A system for probabilistically determining an individual'"'"'s sleep stage, comprising:

a memory storing instructions; and

one or more processors configured to execute the instructions to perform one or more operations, the operations comprising;

receiving a set of signals from a set of sensors worn by the individual, the set of signals including a photoplethysmographic (PPG) signal and an accelerometer signal;

dividing the PPG signal into a set of equally-timed segments if the received PPG signal comprises segments having unequal time durations;

determining a beat interval associated with each segment, the beat interval reflecting an elapsed time between successive heartbeats;

sampling the set of beat intervals to generate an interval signal;

generating a set of signal features based on the interval signal and the accelerometer signal, the set of signal features including a spectrogram of the interval signal; and

determining a sleep stage for the individual by operating on the set of signal features using a sleep stage classifier included in a learning library, wherein the sleep stage classifier comprises a set of functions defining a likelihood that the individual is in the sleep stage based on the set of signal features.

11. The system of claim 10, wherein sampling the set of beat intervals comprises sampling using the frequency of the equally-timed segments.

12. The system of claim 10, wherein sampling the set of beat intervals comprises resampling using a frequency different from the frequency of the equally-time segments.

13. The system of claim 10, wherein the set of functions includes signal weights, and wherein each signal weight is based on a correlation between a feature in the set of signal features and a known sleep stage.

14. The system of claim 13, wherein the operations further comprise:

sampling the accelerometer signal at a first sampling rate;

differentiating the sampled accelerometer signal over each channel associated with the accelerometer signal; and

averaging a magnitude of the differentiated accelerometer signals over an accelerometer time window to generate a time-averaged motion signal;

wherein the set of signal features further comprises the time-averaged motion signal such that the sleep stage is further based on the time-averaged motion signal.

15. The system of claim 13, wherein the set of signal features further comprise an average pulse rate derived from the PPG signal sampled at a second sampling rate and a sum of the absolute energy in each of a plurality of frequency bands in the spectrogram of the interval signal.

16. The system of claim 15, wherein the sleep stage is determined from among a set of potential sleep stages, wherein the set of potential sleep stages comprises an awake stage, a non-REM stage, and a REM stage, and wherein the operations further comprise determining a sleep stage likelihood for each of the potential sleep stages based on the set of signal features and the set of signal weights in the sleep stage classifier.

17. The system of claim 16, wherein determining the sleep stage further comprises:

identifying a most likely sleep stage based on a highest sleep stage likelihood among a set of sleep stage likelihoods including each sleep stage likelihood associated with a potential sleep stage; and

determining whether the highest sleep stage likelihood exceeds a preset stage likelihood threshold.

18. The system of claim 16, wherein the learning library comprises data from among a plurality of sessions for a plurality of individuals, and wherein the set of signal weights is based on normalizing each feature in the set of features over each session for each of the plurality of individuals.

19. A tangible, non-transitory computer-readable medium storing instructions, that, when executed by at least one processor, cause the at least one processor to perform a method for probabilistically determining an individual'"'"'s sleep stage, the method comprising:

receiving a set of signals from a set of sensors worn by the individual, the set of signals including a photoplethysmographic (PPG) signal and an accelerometer signal;

dividing the PPG signal into a set of equally-timed segments if the received PPG signal comprises segments having unequal time durations;

determining a beat interval associated with each segment, the beat interval reflecting an elapsed time between successive heartbeats;

sampling the set of beat intervals to generate an interval signal;

generating a set of signal features based on the interval signal and the accelerometer signal, the set of signal features including a spectrogram of the interval signal; and

determining a sleep stage for the individual by operating on the set of signal features using a sleep stage classifier included in a learning library, wherein the sleep stage classifier comprises a set of functions defining a likelihood that the individual is in the sleep stage based on the set of signal features.

20. The computer-readable medium of claim 19, wherein sampling the set of beat intervals comprises sampling using the frequency of the equally-timed segments.

21. The computer-readable medium of claim 19, wherein sampling the set of beat intervals comprises resampling using a frequency different from the frequency of the equally-time segments.

22. The computer-readable medium of claim 19, wherein the set of functions includes signal weights, each signal weight being based on a correlation between a feature in the set of signal features and a known sleep stage, wherein the learning library comprises data from among a plurality of sessions for a plurality of individuals, wherein the set of signal weights is based on normalizing each feature in the set of features over each session for each of the plurality of individuals, and wherein normalizing the spectrogram of the interval signal is based on the median of the spectrogram and not the mean.

23. The computer-readable medium of claim 22, wherein the method further comprises:

sampling the accelerometer signal at a first sampling rate;

differentiating the sampled accelerometer signal over each channel associated with the accelerometer signal; and

averaging a magnitude of the differentiated accelerometer signals over an accelerometer time window to generate a time-averaged motion signal;

wherein the set of signal features further comprises the time-averaged motion signal such that the sleep stage is further based on the time-averaged motion signal.

24. The computer-readable medium of claim 22, wherein the set of signal features further comprise an average pulse rate derived from the PPG signal sampled at a second sampling rate and a sum of the absolute energy in each of a plurality of frequency bands in the spectrogram of the interval signal.

25. The computer-readable medium of claim 24, wherein the sleep stage is determined from among a set of potential sleep stages, wherein the set of potential sleep stages comprises an awake stage, a non-REM stage, and a REM stage, and wherein the method further comprises determining a sleep stage likelihood for each of the potential sleep stages based on the set of signal features and the set of signal weights in the sleep stage classifier.

26. The computer-readable medium of claim 25, wherein determining the sleep stage further comprises:

identifying a most likely sleep stage based on a highest sleep stage likelihood among a set of sleep stage likelihoods including each sleep stage likelihood associated with a potential sleep stage; and

determining whether the highest sleep stage likelihood exceeds a preset stage likelihood threshold.