Objective To develop a predictive model for microsatellite instability (MSI) status in colorectal cancer based on contrast-enhanced CT radiomics combined with deep learning methods. Methods A total of 211 patients who underwent radical resection for colorectal cancer and were examined by immunohistochemistry (IHC) at Shanghai General Hospital between January 2016 and December 2023 were retrospectively collected. Among the 211 patients, there were 130 with microsatellite stable (MSS) and 81 with microsatellite instability (MSI). The patients were randomly divided into a training group and a validation groups at 7 ∶ 3 ratio. The training group included 57 patients with MSI and 91 patients with MSS. The validation group included 24 patients with MSI and 39 patients with MSS. Three-dimensional volumes of interest (VOIs) of the tumors were manually delineated on venous phase contrast-enhanced CT images. A total of 1458 radiomics features were extracted using PyRadiomics, and 512 deep learning features were obtained using ResNet50. These features were then combined to generate fusion features. Feature selection was performed using intraclass correlation coefficients, Pearson correlation analysis and LASSO regression. Predictive models based on radiomics, deep learning and fusion features were constructed separately using five machine learning algorithms, including support vector machine ( SVM ), logistic regression (LR) , random forest (RF), adaptive boosting(adaboost) and K-nearest neighbors(KNN) Model performance was evaluated by the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA). Results A total of 25 key features were ultimately selected, including 15 radiomics features and 10 deep learning features. In the validation group, the optimal model based on radiomics features was the KNN model. Its AUC was 0. 761 (95% CI = 0. 603 ~ 0． 877 ). The optimal model based on deep learning features was the RF model. Its AUC was 0． 772 (95% CI = 0． 639 ~ 0． 886). The optimal model based on fusion features was the AdaBoost model. Its AUC was 0. 862 (95% CI = 0． 756 ~ 0． 949). Conclusions The combined model based on contrast-enhanced CT radiomics and deep learning features shows high diagnostic value in predicting the MSI status of colorectal cancer. It is expected to support personalized immunotherapy decision-making.