Identification Of Sleep/Wake Stages In Actigraphy Data Utilising Gradient Boosting Algorithm

Mikulec, Marek

Identification Of Sleep/Wake Stages In Actigraphy Data Utilising Gradient Boosting Algorithm

Files

270_EEICT_2021_2.pdf(1.13 MB)

Date

2021

Authors

Mikulec, Marek

Publisher

Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií

Altmetrics

Abstract

Sleep disorders are early markers of various serious diseases that can be treated moreeffectively when diagnosed in their prodromal stage. Actigraphy is a noninvasive sleep monitoringmethod for the detection of sleep patterns and determination of sleep parameters that could support thediagnosis of these disorders. This study aims to compare a newly proposed actigraphy-based methodof sleep/wake detection with a conventional one in terms of consistency with a polysomnography(PSG) reference. 55 recordings (acquired in 28 subjects) of actigraphy and PSG were modelled by aheuristics-based method and by a new approach utilising a gradient boosting algorithm. In addition,another database (22 subjects, 150 recordings) was used to compare scores of the new method withdata reported in sleep diaries. The proposed method achieves 89% accuracy and Mathews correlationcoefficient equal to 0.75 when compared to the polysomnography reference. Such results outperformthe ones provided by the heuristic technique. The newly proposed method has good consistency withthe PSG reference, thus being a good alternative to the golden standard in sleep disorders assessment,especially in decentralised clinical trials.

Keywords

actigraphy, machine learning, polysomnography, sleep, sleep diary

Citation

Proceedings II of the 27st Conference STUDENT EEICT 2021: Selected Papers. s. 270-274. ISBN 978-80-214-5943-4
https://conf.feec.vutbr.cz/eeict/index/pages/view/ke_stazeni

Document type

Peer-reviewed

Document version

Published version

Language of document

en

Document licence