A Computational Approach towards Improving the Accuracy of Exercise Stress Test in Coronary Artery Disease Diagnosis

Abstract

Coronary artery disease (CAD) has probably a ected human beings throughout history, and now it has emerged as a leading cause of death in worldwide. Because of its prevalence and acuteness, several initial screening tests have been designed to perform timely and accurate diagnostic evaluation. Among these screening tests, Exercise Stress Test (EST) which captures the exercise-induced changes in the electrocardiogram has been used to diagnose CAD for almost a century. Over the past decade, however, the role of EST has become controversial because of its limited diagnostic accuracy in terms of low sensitivity and speci city. In this study we propose a combined multi-classi er model to improve the accuracy of EST interpretations in order to determine the presence of CAD with considerably high sensitivity and speci city. This combined classi er model is consisted of two distinct base classi ers where Multilayer Perceptron (MLP) neural network and Decision Tree (DT) algorithms were chosen based on their performance and diversity. As an e ective approach of combining classi ers, the Average Probability-based Combination (APC) rules has been adapted to combine these base classi ers and settle on the nal decision. Finally the overall performances of the combined model as well as the base classi ers were evaluated separately using 10-fold cross validation technique on a study population of 210 patients which de ned by 14 signi cant features. The experimental results indicated that the performance of the combined classi er can achieve 89.52% classi cation accuracy on average, which is 5.23% better than that of the MLP base classi er and 2.38% better than the DT base classi er. Moreover, the performance of combined classi er measured as the area under the receiver operating characteristic (ROC) curve was 95.3% and it was greater than the values provided by any base classi er individually. In conclusion, the results suggest that the diagnostic accuracy of EST in detection of CAD can be further improved up to sensitivity of 89.56% and speci city of 90.62% through the proposed multi-classi er approach.