Coronary artery disease detection based on photoplethysmography via Gramian angular field transformation and deep learning

Sudden cardiac death (SCD) remains one of the leading causes of mortality worldwide, with coronary artery disease (CAD) as its predominant underlying condition. However, noninvasive and accessible screening approaches for CAD are still limited. This study aims to develop and evaluate a photoplethysmography (PPG)‐based method for CAD detection using a two‐dimensional Gramian angular field (GAF) transformation combined with deep learning. We enrolled 89 patients with CAD and 70 healthy controls and converted their PPG signals into two GAF representations—Gramian angular summation field (GASF) and Gramian angular difference field. The GASF representation, which preserves both magnitude and phase relationships within the PPG waveform, was found to provide superior discriminative capability. Using GASF as input, the proposed SE‐ResNet model achieved an accuracy of 92.43% (95% CI: 91.51–93.36), outperforming prior work that reported 83.8% accuracy (95% CI: 82.2–85.3). These results demonstrate that the GAF transformation enhances CAD detection by encoding the temporal–phase dynamics of PPG signals, which are often overlooked in conventional one‐dimensional analyses. The proposed GASF‐SE‐ResNet framework therefore shows strong potential as a noninvasive low‐cost tool for CAD screening and SCD risk reduction.