RAFNet: Restricted attention fusion network for sleep apnea detection

Area of Science:

• Biomedical Engineering
• Cardiology
• Sleep Medicine

Background:

• Sleep apnea (SA) is a prevalent sleep disorder with severe health consequences.
• Portable devices using physiological signals aid SA detection, but performance is limited by signal complexity.
• Electrocardiogram (ECG) signals offer a viable, easily obtainable data source for SA monitoring.

Purpose of the Study:

• To develop an advanced method for sleep apnea detection using single-lead ECG signals.
• To introduce a novel deep learning model, RAFNet, for improved SA detection accuracy.
• To leverage temporal and morphological features from ECG segments for robust SA identification.

Main Methods:

• Proposed a Restricted Attention Fusion Network (RAFNet) for SA detection from ECG.
• Utilized five-minute ECG segments (target, pre-, and post-adjacent) as input.
• Implemented cascaded morphological and temporal attention mechanisms to focus on relevant features and reduce noise.
• Employed channel-wise stacking for feature fusion.

Main Results:

• RAFNet demonstrated significantly improved SA detection performance on public (Apnea-ECG) and clinical (FAH-ECG) datasets.
• The proposed attention mechanism effectively learned discriminative features while suppressing redundant information.
• Achieved competitive and superior results compared to existing state-of-the-art methods.

Conclusions:

• RAFNet offers a powerful and accurate approach for sleep apnea detection using single-lead ECG.
• The attention fusion strategy effectively addresses the complexity and time-variability of physiological signals.
• This method holds promise for enhancing SA diagnosis in portable monitoring applications.