An Enhanced Transformer BiLSTM model for Sleep Apnea Diagnosis based on Single Lead ECG Signal Input

Sleep is an essential biological process vital to cognitive function, immune regulation, and overall physiological balance.Disruptions in sleep, such as sleep apnea, represent a significant public health concern, often leading to cardiovascular complications, daytime fatigue, and impaired quality of life.Traditional diagnostic methods like polysomnography, though accurate, are costly, labor-intensive, and limited in accessibility.Existing machine learning (ML) and deep learning (DL) approaches have attempted to automate sleep apnea detection using physiological signals such as ECG; however, these methods frequently encounter limitations such as inadequate temporal context modeling, high computational overhead, or suboptimal performance on unseen data.In response to these challenges, the present study introduces an innovative hybrid architecture that integrates Transformer encoders and Bidirectional Long Short-Term Memory (BiLSTM) networks for accurate and efficient classification of sleep apnea events.The dataset used for training and evaluation was obtained from Kaggle, based on the Apnea-ECG corpus, a widely accepted benchmark in sleep research.A comprehensive preprocessing pipeline-including normalization, label encoding, feature scaling, and reshaping-was employed to ensure the data's suitability for deep neural computation.The Transformer module leverages multi-head self-attention to capture global dependencies, while the BiLSTM extracts sequential patterns, ensuring robust temporal modeling.The concatenated feature vector is passed through a dense sigmoid classifier for binary classification.The proposed model achieves an accuracy of 98.12%, precision, recall, and F1-score of 0.98, and a near-perfect ROC-AUC, clearly outperforming existing baseline methods.The results demonstrate the model's potential for real-time, non-invasive sleep apnea screening, providing a valuable contribution to intelligent healthcare systems and early detection frameworks.