Predicting Coronary Heart Disease from Sleep-Stage-Specific Electrocardiograms Using Convolutional Neural Networks

Coronary heart disease (CHD) is one of the leading causes of death globally, underscoring the importance of heart health monitoring and early prediction. This study explores utilizing deep convolutional neural networks (CNNs) to predict CHD by analyzing electrocardiogram (ECG) signals collected during overnight sleep. We aimed to compare the predictive value of ECG recordings from different sleep stages: Wake, N1, N2, N3/N4, and REM. The study utilized the Sleep Heart Health Study (SHHS) dataset to train a CNN-based classifier on ECG signals of two different lengths, 30 and 120 seconds, from each sleep stage. The proposed predictive model performed more efficiently when using ECG data from the Wake and N3/N4 stages compared to other stages. The model trained using 30-second ECG during the Wake stage achieved the highest accuracy ($\mathbf{7 5 . 0 \%)}$). The results suggest that analyzing sleep-stage-specific ECG Can potentially serve as a novel approach for early CHD prediction.