Non-invasive detection of coronary artery disease from photoplethysmograph using lumped parameter modelling

Abstract

Coronary artery disease (CAD) is one of the fatal diseases among various cardiovascular diseases (CVDs). Early diagnosis of this cardiac condition is the need of the hour for saving patients’ lives. Coronary angiography is the gold standard clinical method for detecting a blockage in the coronary artery. However, it is an invasive and costly procedure. Due to these limitations, it cannot be used for preventive screening of the general population. In this study, an intelligent predictive model is proposed to diagnose CAD via lumped parameter modelling (LPM) of the human heart. Only non-invasive measurements are used to get a subject-specific CVS model. Here, the clinical data related to the subject and the derived CVS parameters are merged to generate a feature set for the classification of CAD and non-CAD subjects. Four different algorithms are used, namely Decision Tree (DT), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF) classifier, to select the best-performed classification model. For the given data set, the RF classifier is found to work best among these in terms of all performance measures. The classification model is built using ten-fold cross-validation (CV) on the training set. The best performed RF classification model gives a ten-fold CV accuracy of (0.96 ± 0.057)%, whereas the sensitivity and specificity over unseen test sets are 1 and 0.85, respectively. Apart from the CAD classification, the proposed method also can estimate subject-specific CVS model parameters truly non-invasively from signals like photoplethysmograph.