Employing novel DeepBiLSTM framework on NVIDIA Jetson Nano edge platform for sleep apnea detection

Sleep Apnea (SA) is a prevalent sleep disorder characterized by repeated interruptions in breathing, leading to significant health risks if left undiagnosed. This study presents a Deep Bidirectional Long Short-Term Memory (DeepBiLSTM) framework for automated detection of SA using single-lead electrocardiogram (ECG) signals. The problem is formulated as a window-based multi-class classification task, categorizing ECG segments into Wake (W), Obstructive Apnea (OA), and Central Apnea (CA). ECG recordings from the MIT-BIH Polysomnographic Database were segmented into fixed-length windows and labeled based on corresponding annotation intervals. To avoid subject-level bias, a record-wise data split strategy was employed, ensuring that data from each subject appears exclusively in either training or testing sets. A wavelet-sequence preprocessing layer based on discrete wavelet transform (DWT) was integrated with a BiLSTM network to capture both time–frequency characteristics and bidirectional temporal dependencies. The proposed model achieved an overall accuracy of 98.51% demonstrating competitive performance compared to existing methods under controlled evaluation settings. Furthermore, the trained model was deployed on an NVIDIA Jetson Nano edge platform, and quantitative metrics including inference latency, memory usage, and power consumption were evaluated to demonstrate feasibility for real-time applications. The results indicate that the proposed framework provides an effective and practical solution for automated sleep apnea detection using ECG signals.