SleepCode: A Novel Attention-Driven Multimodal Framework for Sleep Staging and Obstructive Apnea Screening

Sleep staging and obstructive sleep apnea (OSA) detection are essential to sleep medicine but remain constrained by the complexity of full polysomnography (PSG) and manual scoring. This article proposes a multimodal monitoring system that integrates EEG and ECG for automated sleep assessment. The system comprises two pipelines: SAGE-EEG, which extracts oscillatory and temporal descriptors through denoising, band decomposition, and spatio-temporal attention; and VITA-ECG, which leverages R-centered sub-epochs, variational mode decomposition, and adaptive feature weighting to capture autonomic dynamics. Both pipelines generate physiologically meaningful features that are encoded using lightweight neural modules and fused for final decision-making. Experiments on the DOD-H/O, ISRUC, and MASS datasets demonstrate robust sleep staging accuracies of 89.2%, 92.4%, and 93.8%, respectively, alongside reliable OSA risk assessment accuracies of 93.5%, 94.7%, and 95.9%. To support practical deployment, the framework is implemented as a clinician-oriented SleepCode Monitor enabling automated sleep staging and integrated OSA screening. While strong performance is observed across heterogeneous datasets and recording configurations, future work will focus on robustness to real-time sensor noise and hardware variability, as well as broader validation in clinical and home-based settings, particularly for patients with severe comorbidities and challenging sensor attachment conditions.