Litcius/Paper detail

Challenges and opportunities of deep learning for wearable-based objective sleep assessment

Bing Zhai, Greg J. Elder, Alan Godfrey

2024npj Digital Medicine21 citationsDOIOpen Access PDF

Abstract

Deep learning (DL, Table 1) , a subset of machine learning (ML), has significantly impacted the field of automated sleep assessment, especially through the analysis of polysomnography (PSG) data. PSG is the most accurate objective sleep measurement method because it simultaneously assesses multiple physiological parameters, including overnight brain activity, and can classify sleep into distinct stages 1 . DL models trained on clinical PSG data have attained performance levels comparable to human experts, providing clinicians with valuable tools for automated and comprehensive sleep stage analysis 2 , 3 , 4 , 5 , across a range of clinical datasets e.g., MESA 6 , SHHS 7 . However, PSG’s suitability for long-term, at-home sleep monitoring is limited due to its intrusive nature. Even headband devices like Dreem™, though less intrusive than traditional PSG technology for brain wave-based sensing, can be cumbersome/uncomfortable during extended wear 8 .

Topics & Concepts

Generalizability theoryWearable computerActigraphyArtificial intelligenceComputer scienceMachine learningSleep (system call)Deep learningField (mathematics)Wearable technologyIntersection (aeronautics)AdaptabilityData scienceHuman–computer interactionPsychologyEngineeringCircadian rhythmAerospace engineeringMathematicsBiologyPure mathematicsEcologyDevelopmental psychologyEmbedded systemNeuroscienceOperating systemObstructive Sleep Apnea ResearchContext-Aware Activity Recognition SystemsSleep and related disorders