SpecExpert-CGC: Multi-Task Co-Learning Framework for Simultaneous Sleep Staging and OSA Event Detection Using EEG and...

Sleep staging and Obstructive Sleep Apnea (OSA) event detection are fundamental aspects of sleep research. However, existing studies have predominantly focused on single-task learning methods, which do not adequately capture the intrinsic relationship between respiratory events and sleep staging. Furthermore, the utilization of EEG signals alone limits the capacity to comprehensively characterize OSA-related physiological dynamics. To address these limitations, this paper proposes SpecExpert-CGC, a novel multi-task co-learning framework based on EEG and ECG signals for simultaneous OSA event detection and sleep staging. Specifically, SpecExpert is designed to integrate depthwise separable convolutions with transformer layers to fully exploit key information from the EEG and ECG spectrograms. Additionally, a Customized Gate Control (CGC) mechanism is employed to effectively fuse shared and task-specific features, further enhancing multi-task learning performance. The experimental results demonstrate that the proposed framework outperforms traditional single-modality approaches, showing potential for improved detection accuracy and robustness.Clinical relevance - The proposed SpecExpert-CGC framework shows potential for improving the accuracy and robustness of OSA event detection and sleep staging through multitask learning with EEG and ECG signals. This may contribute to advancing automated sleep analysis and supporting the early identification and management of sleep disorders.