Semi-Supervised Sleep Stage Classi cation with iBand+

Keywords
Loading...
Thumbnail Image
Issue Date
2022-03-01
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The scale, depth, and implications of many people's poor sleep health have considerable detrimental e ects on society. Unfortunately, sleep studies to help patients or to gain insight into sleep health are heavily limited by the required help of a clinician or researcher to handle the electroencephalography (EEG) system and to interpret the results. iBand+ is a new EEG system that may overcome this limitation as it can be used independently by the user. It requires an integrated automatic sleep tracking system with the performance of manual classi cation by a professional. This project aims to develop a better automatic sleep stage classi er for iBand+ than the currently used random forest classi er, by combining TinySleepNet with recent advancements in semi-supervised learning, which separately have been shown to achieve state-of-the-art performance on the Sleep-EDF dataset. New iBand+ data acquisition experiments are set up to train and evaluate the investigated models. We show that TinySleepNet outperforms the currently used classi er on Sleep-EDF and that both classi ers achieve a similar performance on old iBand+ data compared to each other. We argue why we expect TinySleepNet to outperform the currently used classi er on the new iBand+ data once the acquisition is nished. Furthermore, semisupervised learning showed signi cant improvements for TinySleepNet on both Sleep-EDF and old iBand+ data. Hence, a big step has been taken towards human-level automatic sleep stage classi cation with iBand+, which will contribute to new insights into sleep health and make sleep analysis accessible to the general public.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen