A Robust Hybrid Deep Learning Framework for Sleep Apnea Detection using 1DCNN And Capsule Network on Single-Lead ECG Signals

Sleep apnea (SA) is a prevalent yet underdiagnosed sleep disorder that causes repeated breathing interruptions during sleep, leading to oxygen desaturation and increasing the risk of cardiovascular, metabolic, and cognitive complications. Existing diagnostic methods, such as polysomnography, are accurate but costly, time-consuming, and unsuitable for large-scale or real-time monitoring. Furthermore, many automated detection approaches either depend on handcrafted features or fail to preserve spatial relationships in physiological signals, limiting their robustness and generalizability. This study addresses these limitations by proposing a hybrid deep learning model that integrates onedimensional Convolutional Neural Networks (1DCNN) for local temporal feature extraction with Capsule Networks (CapsNet) for spatial relationship preservation through dynamic routing. Using a balanced single-lead ECG sleep apnea dataset from Kaggle, the model achieved $96.3 \%$ accuracy, $95.9 \%$ precision, $\mathbf{9 5. 7 \%}$ recall, and a $\mathbf{9 5. 8 \%} \mathbf{F 1}$-score, outperforming several state-of-the-art methods. These results demonstrate that the proposed 1DCNN-CapsNet framework provides a robust, noninvasive, and computationally efficient solution for early detection and continuous monitoring of sleep apnea, with strong potential for integration into wearable and remote healthcare systems.