Prediction of Incident Atrial Fibrillation in Population with Ischemic Heart Disease Using Machine Learning with Radiomics and ECG Markers

Ischemic heart disease (IHD) is the main cause of death globally. The coexistence of IHD with atrial fibrillation (AF) can result in a reduced lifespan and severe disabilities. Despite the significant impact of AF on individuals with IHD, the underlying AF susceptibility mechanisms in IHD remain poorly understood. In this work, we propose machine learning (ML) techniques with CMR radiomics to detect incident AF among the IHD population. We used 12-leads Electrocardiograms as a reference, the most common tool for AF diagnosis. The best results were obtained using radiomics with Logistic Regression achieving an AUC of 0.72. Additionally, the rich phenotypic characterization of CMR imaging alterations may offer novel insights into differences in the cardiovascular disease patterns. The shape and textural features of the left atrium in end diastole were the most predominant markers. Our findings demonstrate the potential of combining CMR radiomics with ML to develop more effective early detection strategies for AF in patients with IHD and increase our understanding of AF susceptibility in IHD.