Non-Invasive Adverse Cardiac Risk Assessment Based on Electrocardiogram, Photoplethysmogram and Their Coupling Characteristics

Adverse cardiac events are a major global health concern, with patients who have coronary artery disease and reduced (left ventricular) ejection fraction (CrEF) at significantly higher risk compared to patients with normal coronary arteries and preserved ejection fraction (NCpEF). This work aims to develop a non-invasive and affordable adverse cardiac events risk assessment toolkit by utilizing time-frequency domain features from discrete wavelet transform (DWT) coefficients of an electrocardiogram (ECG) and photoplethysmogram (PPG), as well as the coupling between ECG and PPG. Three popular meta-heuristic feature selection algorithms: binary bat algorithm (BBA), binary cuckoo search (BCS), and binary grey wolf algorithm (BGW) are considered to identify significant ECG and PPG DWT features and ECG-PPG coupling features. BGW outperformed BBA and BCS by yielding better testing accuracy with fewer selected ECG and PPG DWT features. All three algorithms identified six significant ECG-PPG coupling features, which showed better performance compared to the significant ECG and PPG DWT feature sets. An extreme gradient boosting (XGBoost) classifier trained on these six features achieved a maximum testing accuracy of 89% in distinguishing CrEF from NCpEF. The proposed CrEF/NCpEF assessment approach using very short-duration (15-second), synchronous Lead II ECG, and peripheral PPG recordings, which are easily accessible with wearable devices, is well-suited for widespread, remote, and preliminary adverse cardiac event risk assessment.