Deep Attention Networks With Multi-Temporal Information Fusion for Sleep Apnea Detection.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Sleep Apnea (SA) is a common sleep disorder with serious health implications.
• Current diagnosis relies on polysomnography (PSG) and manual scoring, which is time-consuming.
• Traditional machine learning for SA detection requires manual feature engineering.

Purpose of the Study:

• To introduce a novel deep learning framework, the Deep Attention Network with Multi-Temporal Information Fusion (DAN-MTIF), for sleep apnea detection.
• To leverage single-lead electrocardiogram (ECG) signals for non-invasive SA diagnosis.
• To compare the performance of DAN-MTIF against classical machine learning methods.

Main Methods:

• Utilized three 1D convolutional neural network (CNN) blocks to extract features from R-R intervals and R-peak amplitudes at varying segment lengths.
• Integrated a multi-head attention module with self-attention to dynamically weight features from different temporal scales.
• Conducted comprehensive experiments comparing DAN-MTIF with benchmark classical machine learning and deep learning approaches.

Main Results:

• DAN-MTIF achieved high performance metrics: 0.9106 accuracy, 0.9396 precision, 0.8470 sensitivity, 0.9588 specificity, and 0.8909 F1-score at the per-segment level.
• The model effectively extracted discriminative features from multi-timescale ECG segments, outperforming single time-scale feature extraction.
• Deep learning methods, including DAN-MTIF, demonstrated superior performance compared to classical machine learning algorithms for SA detection.

Conclusions:

• The proposed DAN-MTIF framework offers a highly accurate and efficient method for sleep apnea detection using single-lead ECG.
• Multi-temporal feature fusion and attention mechanisms are crucial for enhancing SA detection performance.
• Deep learning approaches represent a significant advancement over traditional methods for automated sleep apnea diagnosis.