Towards an Explainable AI-based Tool to Predict the Presence of Obstructive Coronary Artery Disease

Abstract

Obstructive coronary artery disease (CAD) is characterized as significant upon detection of stenosis of coronary artery diameter. In this paper, we adapt Artificial Intelligence (AI)-based predictive models to accurately estimate the pretest likelihood of obstructive CAD on coronary computed tomography angiography (CCTA) in patients with suspected CAD. In doing so, we use patients' objective results and variables extracted from the screening procedure in combination with demographics, medical history, social history, and other medical data. We use a dataset consisting of 77 patients and we apply a number of alternative Machine Learning (ML) algorithms to predict coronary artery stenosis severity . The ensemble voting model showed the best results across all performance metrics with an area under curve (AUC) of approximately 0.88. We also attempt to provide the clinicians with an explanation of the prediction as to make it more trustworthy.