Coronary artery disease (CAD) is one of the leading causes of mortality worldwide, and coronary artery bypass grafting (CABG) is recognized as the standard treatment for patients with multivessel disease. Accurate prediction of postoperative outcomes in these patients is of substantial clinical importance. In this study, we investigated a novel machine learning and deep learning framework that integrates multimodal data—including coronary angiography images, two-stage blood test results, and clinical variables—to predict postoperative survival and key biochemical outcomes (BUN, SGOT, and SGPT). The dataset comprised 152 patients who underwent CABG at two tertiary cardiac centers in Ahvaz over a four-year period. Deep features were extracted from coronary angiography images using a ResNet-18 architecture and combined with biomarker and clinical data within machine learning pipelines. To reduce dimensionality and improve generalizability, feature selection techniques—including SelectKBest, SelectFromModel with Random Forest, Lasso, and ExtraTrees—were applied, along with a range of classifiers such as logistic regression, support vector classifier (SVC), k-nearest neighbors (KNN), XGBoost, LightGBM, and ensemble methods. Model performance was evaluated using five-fold cross-validation, an independent test set, and standard metrics including accuracy, precision, recall, F1-score, and AUC. The results demonstrated that Lasso-based approaches, particularly when combined with logistic regression and SVC, achieved superior performance, yielding a cross-validated accuracy of 97.14% for survival prediction and accuracies of 93.33%, 93.33%, and 99.05% for BUN, SGOT, and SGPT, respectively. Comparisons with conventional models indicated that multimodal data integration leads to a substantial improvement in predictive accuracy.