Abstract: Objective: This study aimed to construct and validate the efficacy of a radiomics model based on super-resolution (SR) deep learning reconstruction for predicting proximal coronary atherosclerotic plaque (CAP) formation in patients with myocardial bridging (MB) using coronary computed tomography angiography (CCTA). Materials and Methods: A total of 382 patients with MB confirmed by CCTA were retrospectively enrolled and divided into case and control groups based on the presence or absence of proximal CAP (n=191 each). Data were randomly divided into training and test sets in an 8∶2 ratio. A generative adversarial network was employed to perform 4× SR reconstruction on normal-resolution (NR) CCTA images. Regions of interest in the pericoronary adipose tissue were delineated on NR and SR images, and 1106 radiomic features were extracted. The most robust features were selected using the intraclass correlation coefficient, U test, Spearman correlation analysis, and least absolute shrinkage and selection operator regression, followed by extra-tree model construction. The predictive performance of the radiomics model was evaluated using the area under the receiver operating characteristic curve (AUC), and the clinical net benefit was determined through decision curve analysis (DCA). Results: The SR model exhibited significantly superior performance in predicting proximal CAP compared to the NR model, with AUCs increasing to 0.914 (vs. 0.783) and 0.857 (vs. 0.722) in the training and test sets, respectively. DCA confirmed the higher net clinical benefits of the SR model. Conclusion: Generative adversarial network-driven SR reconstruction significantly enhanced radiomic feature characterization. The SR model outperformed the NR model in predicting proximal CAP in MB with higher net benefit, warranting clinical promotion.