Temporal Fusion Transformer Based Temporal Convolutional Network Model for Sleep Apnea Detection Using Single-Lead ECG Signals

Sleep apnea (SA) is a sleep disorder which is characterized by intermittent interruptions in breathing and airflow throughout a sleep cycle. Traditional techniques such as polysomnography (PSG), remain costly, labor-intensive, and require overnight monitoring in specialized labs, which limits its accessibility and scalability. For continuous sleep apnea diagnosis, deep learning models offer a scalable, non-invasive, and economical alternative to multi-sensor PSG monitoring which paves the way for integration into wearable IoT devices for real-time health monitoring. To address the limitations posed by traditional multi-sensor techniques, a novel framework based on Deep Learning techniques is proposed: Temporal Fusion Transformer-based Temporal Convolutional Network (TFT-TCN), designed for the detection of obstructive sleep apnea (OSA) with the use of single-lead ECG signals. The TCN module uses dilated convolutions to extract temporal features efficiently whereas the TFT module employs Gated Residual Network (GRN), Adaptive Fusion and Multi-Head Attention to compute attention scores and fuse the feature vectors. Overall, the proposed TCN-TFT model achieved a noteworthy Accuracy of 89.55 %, FI-Score of 89.3% and AUC score of 0.95.