Prediction of Myocardial Function after Coronary Artery Bypass Graft using Cardiac Magnetic Resonance Imaging Radiomics

Abstract

The purpose of the present study was to predict myocardial function improvement in patients after coronary artery bypass grafting (CABG) using cardiac MRI with late gadolinium enhancement (LGE-CMR)-based radiomics features and machine learning algorithms. Overall, 43 patients who had a visible scar on short-axis series of LGE-CMR images and were candidates for CABG surgery were included. Three months after surgery, echocardiography was performed for all candidates in order to evaluate the post-operation LVEF. We used radiomics methods to prepare and interpret preoperative MR images. Bagging Random Forests (BRF) and Recursive Partitioning were used as feature selectors and classifiers. RPROC curve analysis was used to determine the performance of our models. According to the results for both feature selection algorithms, the "shape Maximum 2D Diameter" feature had the highest importance value in the classification of myocardial function. The BRF model achieved an area under the ROC curve of 0.724, while RP achieved a value of 0.671. The results of the Bootstrap test for comparison of two correlated ROC curves did not show a significant difference between the two models (p-value=0.859). The results of this study showed that machine learning algorithms can provide useful results towards improving myocardial function in patients after CABG. In this study, BRF provided more accurate results in predicting myocardial function.