Abstract: Objective To establish a multi-component machine learning model based on enhanced CT radiomics features and clinical features and compare the performance of the preoperative prediction of Mismatch Repair(MMR)typing model in colorectal cancer(CRC)patients. Methods The clinical and imaging data of 120 colorectal cancer patients from the First Affiliated Hospital of Hebei North University were retrospectively collected, and they were randomly divided into training group and validation group according to 7∶3. The Region of Interest(ROI)was drawn during the portal vein phase of enhanced CT, and the radiomics features were extracted to select the optimal set. Clinical data were analysed using statistical methods. The area under the receiver operating characteristic curve(AUC), sensitivity, accuracy, specificity and FI score were used to evaluate the diagnostic performance of the model, and the calibration and decision curves were plotted. Results The performance of the radiomics features-clinical parameters combined model(hereinafter referred to as the radiomics-clinical combined model, established by combining radiomics features with clinical parameters in 3 algorithms was higher than that of other models. Among them, the logistic regression algorithm in the radiomicsinical combined model showed high performance, with an AUC value of 0.93(95% CI=0.89~0.96)in the validation set, an AUC value of 0.91(95% CI=0.83~0.98)the support vector machine algorithm, and an AUC value of 0.91(95% CI=0.73~0.97)for the random forest algorithm. Conclusion The image-proximity joint model based on enhanced CT images can effectively identify the MMR typing of CRC.

Key words: colorectal cancer, mismatch repair systems, radiomics, machine learning, logistic regression