A Domain-Based, Adaptive, Multi-Scale, Inter-Subject Sleep Stage Classification Network

Sleep stage classification is of great importance in sleep analysis, which provides information for the diagnosis and monitoring sleep-related conditions. To accurately analyze structure under comfortable conditions, many studies have applied deep learning to staging based on single-lead electrocardiograms (ECGs). However, there still room improvement inter-subject classification. In this paper, we propose an end-to-end, multi-scale, subject-adaptive network that improves performance model according architecture, training method, loss calculation. our investigation, a multi-scale residual feature encoder extracted various details support extraction ECGs different situations. After taking domain shift caused by individual differences acquisition conditions into consideration, introduced domain-aligning layer confuse domain. Moreover, enhance model, multi-class focal was used reduce negative impact class imbalance sequence prediction added task assist judging stages. The evaluated public test datasets SHHS2, SHHS1, MESA, obtained mean accuracies (Kappa) 0.849 (0.837), 0.827 (0.790), 0.868 (0.840) awake/light sleep/deep sleep/REM classification, confirms improved solution compared baseline. also performed outstandingly cross-dataset testing. Hence, article makes valuable contributions toward improving reliability staging.