Key result
AdaBoost classifier using synchronized ECG and PPG signals distinguishes CAD severity with an 0.84 F1-score.
Why the study?
There is a growing interest in developing interpretable and non-invasive approaches for coronary artery disease screening with severity assessment.
Does an automated framework using synchronously acquired Lead II ECG and PPG signals accurately assess CAD severity in patients undergoing coronary angiography?
Observational (n=230)
No
Does an automated framework using synchronously acquired Lead II ECG and PPG signals accurately assess CAD severity in patients undergoing coronary angiography?
Effect estimate: Average F1-score of 0.84
An interpretable machine learning framework using synchronously acquired short-duration Lead II ECG and PPG signals can accurately distinguish between severe, minor, and non-CAD patients.
May enable non-invasive CAD severity stratification; leaves open prospective validation before clinical adoption.
There is a growing interest in developing interpretable and non-invasive approaches for coronary artery disease (CAD) screening with severity assessment. This work presents an end-to-end patient-level framework for CAD severity assessment using short-duration synchronously acquired Lead II electrocardiogram (ECG) and photoplethysmogram (PPG) signals. The proposed framework integrates signal quality assessment, nonlinear cardiac-hemodynamic coupling analysis, and interpretable machine learning to distinguish severe CAD (sCAD), minor CAD (mCAD), and non-CAD (nCAD) patients. Novel morphological event-independent signal quality indices are introduced using bounded reconstructed phase space and modulation spectral representations of ECG and PPG signals, eliminating the need for accurate fiducial point detection. Nonlinear ECG-PPG coupling characteristics associated with CAD severity are then characterized using multifractal detrended moving average analysis, multifractal detrended moving average cross-correlation analysis, cross-recurrence quantification analysis, and cross-recurrence network measures. The framework was evaluated on an in-house dataset comprising 230 patients who underwent coronary angiography. Three ensemble classifiers, including extreme gradient boosting (XGBoost), balanced random forest (BRF), and adaptive boosting (AdaBoost), were evaluated and compared. The AdaBoost classifier achieved the best average F1-score of 0.84 for three-class CAD severity assessment, with AUROC values of 0.89 and 0.95 for distinguishing sCAD and mCAD from nCAD patients, respectively. Furthermore, to improve model interpretability, counterfactual explanation-based analysis identified sensitive and stable features based on the magnitude of perturbation required to change CAD severity predictions. The proposed framework with competitive state-of-the-art performance demonstrates the potential of synchronously acquired short-duration Lead II ECG-PPG signals to assist cardiologists in automated CAD severity assessment.
No takes yet. Share an insight, caveat, or question.
Jahnavi et al. (2026) conducted an observational in Coronary artery disease (n=230). Synchronously acquired Lead II ECG and PPG signals with AdaBoost classifier vs. Coronary angiography was evaluated on Three-class CAD severity assessment (sCAD, mCAD, nCAD) (Average F1-score of 0.84). An AdaBoost classifier using synchronously acquired short-duration Lead II ECG and PPG signals achieved an average F1-score of 0.84 for distinguishing severe, minor, and non-coronary artery disease.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: