Objective To investigate the predictive value of multimodal magnetic resonance imaging (MRI) based radiomics model for microsatellite instability (MSI) of rectal cancer.

Methods The retrospective cohort study was conducted. The clinicopathological data of 117 patients with rectal cancer who were admitted to 2 medical centers, including 74 in Ningbo Urology & Nephrology Hospital and 43 in the First Affiliated Hospital of Zhejiang University School of Medicine, from January 2020 to December 2022 were collected. There were 73 males and 44 females, aged (63±5)years. Based on random number table, all 117 patients were divided into the training dataset of 70 cases and the test dataset of 47 cases with a ratio of 7:3. All patients underwent pelvic MRI exami-nation. Observation indicators: (1) construction of radiomics prediction model and analysis of charac-teristics; (2) analysis of factors influencing MSI of rectal cancer in the training dataset; (3) construc-tion and evaluation of the prediction model for MSI of rectal cancer. Measurement data with normal distribution were represented as Mean±SD, and comparison between groups was conducted using the t test. Measurement data with skewed distribution were represented as M(Q₁,Q₃), and compari-son between groups was conducted using the Mann‐Whitney U test. Count data were described as absolute numbers, and comparison between groups was conducted using the chi‐square test. Univariate analysis was conducted using the one way ANOVA and multivariate analysis was conducted using the Logistic regression model with forward method. The receiver operating characteristic curve was drawn, and the area under the curve (AUC), decision curve, calibration curve and Delong test were used to evaluate the predictive ability of prediction model.

Results (1) Construction of radiomics prediction model and analysis of characteristics. Five thousand five hundred and eighty radiomics features were finally extracted from the 117 patients. Based on the feature selection using the maximum correlation minimum redundancy method, and the least absolute shrinkage and selection operator fitting algorithm, 9 radiomics features were finally selected. The radiomics prediction model was constructed based on calculation of the radiomics score. (2) Analysis of factors influencing MSI of rectal cancer in the training dataset. Results of multivariate analysis showed that platelet count was an independent influencing factor for MSI of rectal cancer odds ratio=1.13, 95% confidence interval (CI) as 1.06-1.21, P<0.05. (3) Construction and evaluation of the prediction model for MSI of rectal cancer. The clinical prediction model and clinical‐radiomics combined prediction model were constructed based on the results of multivariate analysis. The AUC of clinical prediction model, radiomics prediction model, clinical⁃radiomics combined prediction model in the training dataset was 0.94 (95%CI as 0.86-0.98), 0.96 (95%CI as 0.88-0.99), 0.99 (95%CI as 0.93-1.00), respectively, with the sensitivity and specificity as 90.7%, 91.2%, 96.9% and 85.0%, 88.9%, 94.3%. Results of Delong test showed that there was a significant difference in the predictive performance between the clinical‐radiomics combined prediction model and the clinical prediction model (Z=2.20, P<0.05), and there was no significant difference between the radiomics prediction model and the clinical‐radiomics combined prediction model or the clinical prediction model (Z=1.94, 0.60, P>0.05). The AUC of clinical prediction model, radiomics prediction model, clinical‐radiomics combined prediction model in the test dataset was 0.97 (95%CI as 0.88-1.00), 0.86 (95%CI as 0.73-0.95), 0.97(95%CI as 0.87-1.00), respectively, with the sensitivity and specificity as 99.3%, 95.8%, 99.3% and 85.7%, 73.9%, 90.5%. Results of Delong test showed that there was a significant difference in the predictive performance between the clinical⁃radiomics combined prediction model and the radiomics prediction model (Z=2.21, P<0.05), and there was no significant difference between the clinical prediction model and the clinical⁃radiomics combined prediction model or the radiomics prediction model (Z=0.17, 1.82, P>0.05). Results of calibration curve showed that clinical prediction model, radiomics prediction model, clinical‐radiomics combined prediction model had good ability in predicting the MSI status of rectal cancer. Results of decision curve showed that compared to clinical prediction model and radiomics prediction model, clinical-radiomics combined prediction model had greatest net gain in predicting the MSI of rectal cancer.

Conclusion The prediction model based on 9 radiomics features after selecting can effectively predict the MSI status of rectal cancer, and the clinical-radiomics combined prediction model has a better prediction efficiency.