Automatic Sleep Staging using Hidden Markov Models
Sleep staging consists of taking data from a polysomnography and labeling the time intervals according to the different possible sleep stages (1, 2, 3, 4 and REM). Sleep staging is an integral part of sleep medicine, as it enables specialists to diagnose different types of sleep disorders.
The process of sleep staging is usually done by hand by a qualified technician, taking anywhere from two to three hours depending on the technician's level of expertise and care. Sleep staging can be viewed as a classification problem. As such, it is one that can bene t from machine learning for a faster and thus less expensive solution.
Researchers have studied automated sleep staging in the past [1], and some have even commercialized automated sleep staging systems [2]. These systems perform classification, however, based on sets of heuristic rules rather than on supervised machine learning techniques.
The particular model we have decided to use are Hidden Markov Models. This model seems suitable because in this case, the hidden states are the sleep stages, while the visible states are the different signal values. Additionally, this model seems suitable for the problem at hand given the time-dependent nature of the sleep stages. That is to say, the probability of an interval corresponding to a given stage, is dependent on the sleep stage of the previous interval. For in stance, a patient is much likelier to transition from stage 3 to stage 4 than he is from stage 3 to stage 2, as humans tend to follow these stages in their given order.
See plan.md and classification.md.
[1] Performance of an Automated Polysomnography Scoring System Versus
Computer-Assisted Manual Scoring, Malhotra et al., SLEEP, 2012
[2] Michele Sleep Scoring https://michelesleepscoring.com/