Ensemble Learning-Based Coronary Artery Disease Detection Using Computer Tomography Images

Abstract

Coronary artery disease (CAD) is the most prevalent form of cardiovascular disease that may result in myocardial infarction. Annually, it leads to millions of fatalities and causes billions of dollars in global economic losses. Limited resources and complexities in interpreting results pose challenges to healthcare centers in implementing deep learning (DL)-based CAD detection models. Ensemble learning (EL) allows developers to build an effective CAD detection model by integrating the outcomes of multiple medical imaging models. In this study, the authors build an EL-based CAD detection model to identify CAD from coronary computer tomography angiography (CCTA) images. They employ a feature engineering technique, including MobileNet V3, CatBoost, and LightGBM models. A random forest (RF) classifier is used to ensemble the outcomes of the CatBoost and LightGBM models. The authors generalize the model using two benchmark datasets. The proposed model achieved an accuracy of 99.7% and 99.6% with limited computational resources. The generalization results highlight the importance of the proposed model’s efficiency in identifying CAD from the CCTA images. Healthcare centers and cardiologists can benefit from the proposed model to identify CAD in the initial stages. The proposed feature engineering can be extended using a liquid neural network model to reduce computational resources.