Time Series and Morphological Feature Extraction for Classifying Coronary Artery Disease from Photoplethysmogram

Abstract

In this paper we propose a feature extraction algorithm for classifying Coronary Artery Disease (CAD) patients from Photoplethysmogram (PPG) signals. Several domain-independent features, representing inherent properties of a time series are explored in our study. These are combined with Heart Rate Variability (HRV) and other popularly used morphological PPG features. A statistical feature selection algorithm, based on Maximal Information Coefficient (MIC) is applied on MIMIC II dataset for ranking and choosing the optimum features. A second hospital dataset of different patient demography is used for performance evaluation. Results show that, Support Vector Machine (SVM) classifier, designed using the selected features yields average sensitivity and specificity of more than 0.8 in identifying CAD patients and also outperforms two recent prior art approaches when applied on the test dataset.