Deep Learning-Based Sleep Apnea Classification with Hybrid 1D-CNN-LSTM and 1D-CNN-GRU Models

Sleep apnea is a common sleep disorder characterized by recurrent interruptions in breathing during sleep, leading to severe health complications such as cardiovascular diseases and reduced quality of life. Early and accurate diagnosis is essential for effectively managing and preventing these risks. This study proposes two hybrid deep learning architectures, 1D-CNN-LSTM and 1D-CNN-GRU, for sleep apnea classification using ECG signals from the PhysioNet Apnea-ECG database. The models leverage R-R intervals and R-wave amplitudes as primary input features, where 1D-CNN extracts spatial features while LSTM/GRU captures temporal dependencies in the data. Experimental results using k-fold cross-validation for robust evaluation, the models demonstrated exceptional performance that 1D-CNN-LSTM achieves an accuracy of 96.49%, sensitivity of 94.68%, specificity of 98.33%, and an F1-score of 0.96, while 1D-CNN-GRU attains an accuracy of 96.88%, sensitivity of 96.42%, specificity of 97.35%, and an F1-score of 0.96. These findings indicate that both architectures effectively classify sleep apnea, with 1D-CNN-GRU demonstrating a marginally better overall performance. The results highlight the potential of hybrid deep learning models in developing robust and efficient sleep apnea detection systems, which could facilitate automated diagnosis and real-time monitoring for improved clinical outcomes.