An apparatus for automatically classifying time intervals into one of a plurality of sleep stages, comprising:
- one or more processors;
- memory coupled with the one or more processors;
- and
- instructions stored in the memory and executable by the one or more processors to cause the apparatus to:
- sample, via one or more sensors of a wearable device, a photoplethysmogram (PPG) signal to determine a pulse waveform of the user, wherein the one or more sensors include at least one optical transmitter and at least one optical receiver, the at least one optical transmitter configured to transmit light to illuminate a region of a user, the at least one optical receiver configured to receive light transmitted from the at least one optical transmitter and generate the PPG signal indicative of an amount of the light received, and the PPG signal indicative of perfusion of blood in the illuminated region;
- determine, via the one or more sensors, physiological data associated with a user, wherein the physiological data is based in part on the pulse waveform of the user, wherein the physiological data comprises temperature data, heart rate data, and heart rate variability data, wherein the physiological data is continuously determined throughout a first sleep period, and wherein the first sleep period comprises a plurality of time intervals;
- modify a baseline circadian rhythm adjustment model in accordance with the physiological data to generate a modified circadian rhythm adjustment model corresponding to the user, wherein the modified circadian rhythm adjustment model comprises a circadian drive component, a homeostatic sleep pressure component, and an elapsed sleep duration component, wherein a temporal starting point for the circadian drive component, the homeostatic sleep pressure component, and the elapsed sleep duration component corresponds to a bedtime of the user;
- determine the bedtime of the user associated with the first sleep period based at least in part on the physiological data;
- identify one or more sensor-independent circadian features representative of a circadian rhythm associated with the user based on the modified circadian rhythm adjustment model, wherein the one or more sensor-independent circadian features are representative of physiological changes in sleep stage distribution of the user throughout the first sleep period;
- assign a first weight to a first probability metric for a first time interval of the plurality of time intervals based at least in part on the modified circadian rhythm adjustment model and a first elapsed time between the bedtime of the user and the first time interval, wherein the first probability metric is associated with a probability of the first time interval corresponding to a deep sleep stage of the plurality of sleep stages;
- assign a second weight to a second probability metric for a second time interval of the plurality of time intervals based at least in part on the modified circadian rhythm adjustment model and a second elapsed time between the bedtime of the user and the second time interval, wherein the second probability metric is associated with a probability of the second time interval corresponding to the deep sleep stage, wherein the first weight is greater than the second weight based at least in part on the first time interval occurring earlier in the first sleep period than the second time interval;
- classify, using a machine learning model trained to include the one or more sensor-independent circadian features modeled by the modified circadian rhythm adjustment model, the first time interval as the deep sleep stage based at least in part on the first weight of the first probability metric;
- and
- cause a graphical user interface of a user device to display an indication of the plurality of time intervals of the first sleep period including at least the deep sleep stage classified using the machine learning model.