Abstract
Sleep Apnea Syndrome (SAS) is a prevalent sleep disorder characterized by intermittent pauses in breathing during sleep. If undiagnosed and untreated, SAS can have significant adverse effects on the human physiological system. Polysomnography (PSG) has been regarded as a gold-standard examination method for diagnosing sleep snoring (sleep apnea-hypopnea syndrome, OSAHS), but is often seen as inconvenient due to its complex operational requirements. This study introduces a novel method for SAS detection using temporal ECG and SPO2 signals via a CNN-RNN based Multimodal Multichannel Transfer Module with Squeeze and Excitation (MMTM-SE). Three hybrid CNN-RNN models were developed to extract features from ECG and SPO2 data. These extracted features were then fused through MMTM-SE structure, so as to enhance the correlation between different modalities and adaptively recalibrate channel features. The proposed method was validated by using the Apnea-ECG database across three deep learning networks. The experimental results show that the proposed approach outperformed existing methods, achieving a highest detection accuracy of 98.9%.
Original language | English |
---|---|
Article number | 107589 |
Journal | Biomedical Signal Processing and Control |
Volume | 104 |
DOIs | |
Publication status | Published - Jun 2025 |
Keywords
- Deep learning
- ECG
- Multimodal fusion
- Sleep Apnea syndrome
- SPO2