Deep learning for multimodal physiological signal based assessment of sleep disordered breathing.

Sleep disordered breathing (SDB) is commonly assessed using polysomnography (PSG), which records multiple physiological signals during sleep. Accurate automated assessment of SDB related breathing abnormalities remains challenging because respiratory events occur within complex and stage dependent sleep dynamics. To address this problem, this study proposes a multimodal physiological signal based deep learning framework for SDB oriented sleep assessment. The proposed framework treats SDB related respiratory event detection as the clinically targeted task and incorporates sleep stage classification as an auxiliary contextual task to enhance temporal interpretation of physiological abnormalities. The framework combines a Sleep Pattern Recognition Network (SPRNet) with an Integrated Diagnostic Strategy for Sleep Assessment (IDSSA). SPRNet integrates electroencephalography (EEG), electrocardiography (ECG), electrooculography (EOG), electromyography (EMG), and respiratory airflow signals through modality specific feature extraction, followed by BiLSTM and attention based temporal modeling. IDSSA further incorporates domain informed sleep transition priors, probabilistic reasoning, and multi objective optimization to improve temporal consistency and robustness under noisy physiological recordings. Sleep stage classification is evaluated on both Sleep EDF and SHHS, while SDB related respiratory event detection is evaluated on SHHS, which provides respiratory event annotations. Experimental results show that the proposed framework achieves competitive performance compared with conventional machine learning and deep learning baselines. These findings indicate that multimodal physiological signal integration and temporally informed joint modeling can provide effective decision support for automated assessment of SDB related respiratory abnormalities. The proposed system is intended for automated screening and assessment support rather than as a replacement for clinical diagnosis by sleep specialists.