Model uncertainty quantification for diagnosis of each main coronary artery stenosis

• Published in: Soft computing (Berlin, Germany) Germany), 2020-07, Vol.24 (13), p.10149-10160
• Main Authors: Alizadehsani, Roohallah, Roshanzamir, Mohamad, Abdar, Moloud, Beykikhoshk, Adham, Zangooei, Mohammad Hossein, Khosravi, Abbas, Nahavandi, Saeid, Tan, Ru San, Acharya, U. Rajendra
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Angiography; Artificial Intelligence; Cardiovascular disease; Classification; Computational Intelligence; Control; Coronary artery disease; Coronary vessels; Data mining; Datasets; Decision trees; Diagnosis; Discriminant analysis; Electrocardiography; Engineering; Feature selection; Genetic algorithms; Hyperplanes; Machine learning; Mathematical Logic and Foundations; Mechatronics; Medical imaging; Methodologies and Application; Methods; Neural networks; Principal components analysis; Robotics; Support vector machines; Uncertainty; Wavelet transforms
• ISSN: 1432-7643; 1433-7479
• DOI: 10.1007/s00500-019-04531-0

One of the main causes of death in the world is coronary artery disease (CAD). CAD occurs when there is stenosis in one or more of the three major coronary arteries: right coronary artery (RCA), left circumflex (LCX) artery, and left anterior descending (LAD) artery. The gold standard or CAD diagnosis is angiography, but it is invasive, costly, and time consuming. Therefore, researchers continually seek new machine learning methods that can screen for CAD non-invasively. For reliable and cost-effective CAD diagnosis, several algorithms have been developed. Most prior studies analyzed the presence or absence of CAD in a dichotomous manner. Herein, we studied the more complex problem of classification of stenosis in individual LAD, LCX, and RCA by applying machine learning algorithms on the Z-Alizadeh Sani dataset that comprised 303 subjects, each with 54 features. In addition, our new methodology is developed to handle model uncertainty in the prediction of individual artery stenosis. It uses the hyperplane distance from a sample and accuracy rate of the classifier during the training phase to enhance its performance. Our results demonstrate high diagnostic performance of the proposed method for diagnosis of stenosis in individual RCA, LCX, and LAD, achieving accuracy rates of 82.67%, 83.67% and 86.43%, respectively. This is the best performance of ML techniques applied to the Z-Alizadeh Sani dataset.