WaveCrossNet: interpretable cross-modal automatic sleep staging for OSA patients

Obstructive sleep apnea (OSA) patients show highly fragmented sleep with frequent stage transitions, challenging existing automatic sleep staging in feature extraction, multimodal correlation modeling, and interpretability. Current methods rely on fixed-threshold denoising and standard positional encoding, suffering from obvious module stacking and a lack of design tailored to OSA's non-stationary signals. To address these issues, this paper proposes WaveCrossNet, an enhanced Transformer-based model for OSA sleep staging. Instead of simple module stacking, WaveCrossNet embeds adaptive wavelet thresholding and dual-channel convolutional neural network into Transformer's end-to-end temporal learning pipeline. First, an energy-adaptive subband thresholding mechanism selectively denoises while preserving transient events such as micro-arousals. Second, position-aware attention with dynamic cross-modal fusion explicitly models complementary patterns of electroencephalography and electrooculography acrossN1, REM and other sleep phases. Third, a hard example mining loss alleviates class imbalance, where minority stages likeN1 account for less than 5% of OSA data. Attention weight visualization links feature correlations to physiological events (e.g. slow eye movements inN1, high-amplitude slow waves inN3), enhancing interpretability. Evaluated on the Sleep-EDFx healthy dataset and 200 Sleep Heart Health Study OSA records with subject-independent splits, WaveCrossNet achieves competitive metrics on healthy subjects and obtains state-of-the-art performance on OSA patients with an overall accuracy of 84.9%. By jointly optimizing denoising, representation learning, and temporal modeling, WaveCrossNet provides an interpretable, end-to-end solution for accurate OSA sleep staging.