Machine Learning for Sleep Stage Classification

Research output: Contribution to conferencesPresentation slidesContributed

Abstract

Introduction
Polysomnography is the gold standard for clinical diagnosis of sleep disorders. Several biosignals, including electroen- cephalogram, electrooculogram, electromyogram and nasal flow, are recorded for a full night and manually evaluated by visual criteria. The evaluation contains a sleep stage for each 30 second epoch and additional event annotations. Since the process is time consuming and subjective, automated classifiers or recommender systems could improve efficiency and objectivity of the evaluation. This contribution gives an overview about machine learning for sleep stage classification from different input signals.
Conclusion
The last decade has shown massive improvements in machine learning in general, and many approaches for automated sleep staging have been published. Some models show classification performance comparable to medical experts when evaluating the polysomnogram. However, machine learning also allows for evaluating signals that lack rules for manual scoring. This led to research into automated sleep staging from cardiac, respiratory and movement signals with good classification results. At the same time, innovative non-contact measurements have been investigated for sleep monitor- ing, again with machine learning approaches for sleep evaluation. By now, there are setups based on e.g. radar or cameras, that show good performance for classifying sleep stages from the distance.

Details

Original languageGerman
Publication statusPublished - 7 Oct 2021
Peer-reviewedNo