A Comprehensive Machine Learning of a Radiomics-Based model for predicting microsatellite instability in right Colon Cancer

Abstract

Objective To develop and validate a non-invasive radiomics-based ML model integrated with clinicopathological features for predicting MSI-H status in right colon cancer, aiming to provide a preoperative decision-making tool for clinical practice. Methods: A total of 247 patients with right colon cancer (43 dMMR and 204 proficient mismatch repair [pMMR]) who underwent radical resection between January 1st 2017 and December 31th2024 were enrolled and randomly divided into training set and test set . Preoperative contrast-enhanced CT images were processed using 3D-Slicer for ROI delineation and radiomic feature extraction. ICC was used to assess interobserver consistency and LASSO was applied for feature selection. LR, RF, SVM and XGBoost were used to construct radiomics models. The RF algorithm was selected to build a joint clinicopathological-radiomics model and patients of left colon cancer were used as the external validation set. Receiver Operating Characteristic (ROC) curves, calibration curves and Decision Curve Analysis (DCA) were used to evaluate diagnostic efficiency. Results: A total of 107 radiomic features were extracted ,17 stable features were retained after ICC filtering (ICC ≥ 0.75) and LASSO with 50% cross-validation. The RF algorithm outperformed other models in the radiomics model with an AUC of 0.98 in the training set and 0.96 in the test set. The joint model integrating RF algorithm and clinicopathological variables (sex, age, tumor long diameter, histological type,pN, pM, pTNM and differentiation degree) achieved the highest predictive performance with AUC of 0.99 (training set) and 0.97 (test set), which were significantly higher than radiomics model and clinical model alone. External validation with left colon cancer data also showed an AUC of 0.81, indicating good generalizability. Calibration curves demonstrated satisfactory probability prediction and Decision Curve Analysis (DCA) confirmed that the joint model provided greater clinical net benefit across the entire threshold probability range. Conclusion: The RF-based joint clinicopathological-radiomics model exhibited excellent performance in predicting dMMR status in right colon cancer, with good generalizability across the entire colon. This non-invasive model can serve as a reliable clinical decision support tool to optimize risk stratification and guide early intervention for patients with right colon cancer.