Abstract
BACKGROUND: KRAS mutations are associated with treatment and prognostic outcomes in colorectal cancer patients.There have been no studies on utilizing the peritumoral images to predict KRAS mutation status in rectal cancer patients. We aim to develop a radiomics model utilizing intratumoral and peritumoral ultrasound images for predicting KRAS mutation status in rectal cancer. METHODS: This study retrospectively included 278 patients with pathologically confirmed rectal cancer following surgery, who were randomly divided into training group (194 cases) and test group (84 cases) at a 7:3 ratio. Radiomic features from both intratumoral and peritumoral regions were extracted from endorectal ultrasound images. A five-step procedure was used to select robust features. Based on these, intratumoral, peritumoral, and combined models were created using logistic regression, support vector machine, and light gradient boosting machine. The predictive accuracy, calibration, and clinical utility of the models were evaluated using receiver operating characteristic curves, calibration curves, and decision curve analysis. The SHapley Additive exPlanation (SHAP) method was used to evaluate the importance of the features in the optimal models. RESULTS: In the test set, the area under the curve of all three combined models exceeded that of the individual intratumoral and peritumoral models. According to area under the curve, decision curve analysis, and calibration curves, combine_LR demonstrated the best performance, with an area under the curve of 0.881. CONCLUSIONS: The ultrasound radiomics model, incorporating both intratumoral and peritumoral features, effectively predicts KRAS status in rectal cancer patients, potentially guiding clinical targeted therapy selection.