Introduction: Coronary Artery Disease (CAD), one of the leading causes of death, is narrowing the walls of the coronary arteries. Angiography is the most accurate but invasive and costly CAD diagnosis method associated with mortality. The aim of this study was to design a computer-based non-invasive CAD diagnosis system.
Methods: In this work, a dataset from Cleveland clinic foundation, containing 303 patients and 20 features, was used. Supervised Fuzzy C-means (SFCM) classification was used to design a classifier for CAD diagnosis. The Generalized Minkowski Metrics (GMM) was used to handle objects containing different measurement scale features. The performance of the SFCM was assessed with/without Statistical Feature Selection (SFS). The weights of the GMM, i.e. the significance of different features,beside other classifier parameters were tuned using Differential Search Algorithm (DSA), and the validity of the proposed classifier was further investigated. The hold-out and 10-fold cross validation were used for the performance assessment.
Result: The average accuracy of the base classifier (SFCM + GMM) was 79% (hold-out validation). It increased to 82% when using SFS. The average accuracy, sensitivity and specificity of the DSA-based classifier were 88%, 86% and 88%, respectively (cross-validation).
Conclusion: The most important features were the number of major vessels colored by fluoroscopy, the family history of CAD, peak exercise systolic blood pressure, maximum exercise heart rate achieved, chest pain type, resting heart rate, Fasting Blood Sugar and gender.This classifier showed substantial agreement with the angiographic results. The hybrid diagnosis system is thus promising. However, it is necessary to improve its reliability.
Published on: Jun 8, 2015 Pages: 6-14