Deep Learning for Coronary Artery Stenosis Localization: Comparative Insights from Electrocardiograms (ECG), Photoplethysmograph (PPG) and Their Fusion

Abstract

Coronary artery stenosis (CAS) is a critical cardiovascular condition that demands accurate localization for effective treatment and improved patient outcomes. This study addresses the challenge of enhancing CAS localization through a comparative analysis of deep learning techniques applied to electrocardiogram (ECG), photoplethysmograph (PPG), and their combined signals. The primary research question centers on whether the fusion of ECG and PPG signals, analyzed through advanced deep learning architectures, can surpass the accuracy of individual modalities in localizing stenosis in the left anterior descending (LAD), left circumflex (LCX), and right coronary arteries (RCA). Using a dataset of 7,165 recordings from CAS patients, three models—CNN, CNN-LSTM, and CNN-LSTM-ATTN—were evaluated. The CNN-LSTM-ATTN model achieved the highest localization accuracy (98.12%) and perfect AUC scores (1.00) across all arteries, demonstrating the efficacy of multimodal signal integration and attention mechanisms. This research highlights the potential of combining ECG and PPG signals for non-invasive CAS diagnostics, offering a significant advancement in real-time clinical applications. However, limitations include the relatively small dataset size and the focus on single-lead ECG and PPG signals, which may affect the generalizability to broader populations. Future studies should explore larger datasets and multi-lead signal integration to further validate the findings.