An intelligent noninvasive model for coronary artery disease detection

Abstract

Coronary artery disease (CAD) is one of the leading causes of death globally. Angiography is one of the benchmarked diagnoses for detection of CAD; however, it is costly, invasive, and requires a high level of technical expertise. This paper discusses a data mining technique that uses noninvasive clinical data to identify CAD cases. The clinical data of 335 subjects were collected at the cardiology department, Indira Gandhi Medical College, Shimla, India, over the period of 2012–2013. Only 48.9% subjects showed coronary stenosis in coronary angiography and were confirmed cases of CAD. A large number of cases (171 out of 335) were found normal after invasive diagnosis. Hence, a requirement of noninvasive technique was felt that could identify CAD cases without going for invasive diagnosis. We applied data mining classification techniques on noninvasive clinical data. The data set is analyzed using a hybrid and novel k-means cluster centroid-based method for missing value imputation and C4.5, NB Tree and multilayer perceptron for modeling to predict CAD patients. The proposed hybrid method increases the accuracy achieved by the basic techniques of classification. This framework is a promising tool for screening CAD and its severity with high probability and low cost.