Objective：To develop a three-dimensional convolutional neural network (3D-CNN) model based on coronary computed tomography angiography (CCTA) for predicting rapid plaque progression (RPP), and to compare its performance against traditional machine learning models and existing advanced methodologies. Methods:This retrospective study analyzed 150 patients who underwent serial CCTA examinations. Following strict alignment of CTA volume data with plaque masks, traditional machine learning models (LASSO, Elastic Net, Random Forest, XGBoost) and a lightweight 3D-CNN model were constructed. RPP was defined as an annualized plaque burden (PB) increase ≥ 1.0%. Model performance was primarily evaluated using the area under the receiver operating characteristic curve (AUC), with SHAP (SHapley Additive exPlanations) employed for model interpretation. Results:Traditional models demonstrated limited discriminatory ability, with AUCs ranging from 0.32 to 0.51. The developed 3D-CNN model achieved an AUC of 0.75 on the independent test set, with a sensitivity of 0.64 and a specificity of 0.88. SHAP analysis revealed that the 3D-CNN focused on internal plaque texture and Hounsfield Unit (HU) distribution patterns, whereas traditional models relied on limited features such as plaque volume.Conclusion:The 3D-CNN model can directly learn deep features associated with RPP from CCTA images. Its performance is significantly superior to traditional models and demonstrates potential comparable to current advanced radiomics and machine learning methods, offering a novel tool for non-invasive identification of high-risk plaques using a single time-point baseline scan.