OSA-Net: A Deep Learning Approach for Obstructive Sleep Apnea Classification Using ECG Scalograms

Obstructive Sleep Apnea (OSA) is a serious sleep-related respiratory disorder, marked by periodic interruptions in breathing, leading to severe cardiovascular complications. Due to the complexity and labor demands of polysomnography (PSG), early diagnosis of OSA remains challenging. To address this, this study proposes an automatic classification approach called the Obstructive Sleep Apnea Network (OSA-Net). This model uses ECG scalograms generated through wavelet transform for automatic OSA detection. OSA-Net effectively captures complex dynamic features while preserving dependencies between local information and global context. Extensive experiments on the Apnea-ECG dataset, using five-fold cross-validation, demonstrate the proposed framework's outstanding performance, achieving an accuracy of 0.910, sensitivity of 0.886, specificity of 0.933, and an F1 score of 0.907. These results highlight the potential of the proposed method in assisting clinical decision-making in OSA diagnosis.