Radiomics‐Based CT Prediction of KRAS/NRAS/BRAF Mutation Status in Colorectal Cancer: A Multicenter Study

ABSTRACT Background Colorectal cancer (CRC) is a common malignancy with high morbidity and mortality. Kirsten rat sarcoma viral oncogene homolog (KRAS), neuroblastoma RAS viral oncogene homolog (NRAS), and B‐Raf proto‐oncogene serine/threonine kinase (BRAF) mutations are key biomarkers for targeted therapy. Radiomics offers a non‐invasive approach to predict genetic alterations using CT imaging. The aim of this study was to develop a predictive model for KRAS/NRAS/BRAF gene mutations in colorectal cancer using radiomic characteristics obtained from computed tomography (CT) imaging. Methods A total of 385 patients diagnosed with colorectal cancer were enrolled in this retrospective multicenter study. Radiomic features were extracted by delineating volumes of interest on venous‐phase CT scans. Feature selection was performed using Pearson correlation analysis and the least absolute shrinkage and selection operator (LASSO) algorithm, and a radiomic model was constructed using a support vector machine. Logistic regression was used to develop the clinical model and the combined clinical‐radiomic model. The predictive performance of the models was evaluated using receiver operating characteristic curve analysis, calibration curve assessment, and decision curve analysis. Additionally, heatmaps and Shapley additive explanation plots were used to enhance the interpretability of the models. Results Post feature selection and dimension reduction, a total of six features were maintained for the construction of the radiomic model. In the training dataset, the radiomic model secured an area under the curve of 0.825 (95% CI: 0.769–0.883) compared with 0.821 (95% CI: 0.735–0.908) for the internal validation dataset and 0.818 (95% CI: 0.724–0.912) for the external test dataset. The levels of N stage and Carbohydrate antigen 199 demonstrated a notable correlation with the presence of KRAS/NRAS/BRAF mutations in the treatment of colorectal cancer ( p < 0.05). When combined, the clinical‐radiomic model exhibited enhanced diagnostic precision over using only radiomic models. Conclusions The results demonstrated a correlation between radiomic attributes from CT scans and KRAS/NRAS/BRAF mutations with an improvement in diagnostic efficacy when integrated with relevant clinical factors. CT scans could be a crucial instrument in assessing the genetic state of tumors in colorectal cancer patients, possibly assisting in the formulation of therapeutic approaches.