Multi‐sequence MRI‐based clinical‐radiomics models for the preoperative prediction of microsatellite instability‐high status in endometrial cancer

Purpose: To assess the efficacy of clinical radiomics models in predicting microsatellite instability-high status in endometrial cancer and to identify patients who may benefit from immunotherapy. Materials and Methods: Two hundred and twenty-two patients with endometrial cancer who were consecutively admitted to Yantai Yuhuangding Hospital between January 2021 and April 2022 were retrospectively recruited, and 64 were excluded. Of the remaining 158 patients, 110 and 48 were randomly divided into the training and test sets, respectively. Regions of interest were delineated, and radiomic features were extracted from dynamic contrast-enhanced T1-weighted, T2-weighted, and apparent diffusion coefficient images. The intraclass correlation coefficients, Spearman correlation analysis, Mann-Whitney U test, and least absolute shrinkage and selection operator (LASSO) algorithm were employed for feature selection in radiomics models' development. Seven clinical risk factors were incorporated into the clinical models. Finally, the clinical-radiomics models integrating clinical risk factors and radiomic features were constructed. Clinical, radiomics, and clinical-radiomics models were developed in the training set using logistic regression (LR), random forest (RF), and support vector machine (SVM). The performance of the models was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analyses (DCA). Results: Four clinical factors (progesterone receptor, tumor suppressor gene p53, diabetes, and carbohydrate antigen 153) and 15 radiomic features were recognized as key predictors of microsatellite instability-high status in endometrial cancer. The clinical-radiomics models developed using the SVM classifier exhibited the best performance in the test set, achieving an area under the curve (AUC) of 0.997, sensitivity of 1.000, specificity of 0.936, and accuracy of 0.952. DCA demonstrated that the SVM-based clinical-radiomics model achieved a greater net clinical benefit than the clinical and radiomics models across threshold probabilities ranging from 0 to 0.405 and 0.585 to 1, respectively. Conclusion: The clinical-radiomics nomogram constructed using the SVM classifier exhibited robust predictive performance for microsatellite instability-high status in endometrial cancer. This nomogram may assist in identifying patients with endometrial cancer who are likely to benefit from immunotherapy, thereby providing a tool for personalized immune management.