Development and validation of magnetic resonance imaging-based radiomics models for preoperative prediction of microsatellite instability in rectal cancer.

Abstract

BackgroundMicrosatellite instability (MSI) is a predictive biomarker for response to chemotherapy and a prognostic biomarker for clinical outcomes of rectal cancer. The purpose of this study was to develop and validate radiomics models for preoperative prediction of the MSI status of rectal cancer based on magnetic resonance (MR) images.MethodsThis study retrospectively recruited 491 rectal cancer patients with pathologically confirmed MSI status. Patients were randomly divided into a training cohort (n=327) and a validation cohort (n=164). The most predictive radiomics features were selected using intraclass correlation coefficient (ICC) analysis, the two-sample t test, and the least absolute shrinkage and selection operator (LASSO) method. XGBoost models were constructed in the training cohort to discriminate the MSI status using clinical factors, radiomics features, or a combined model incorporating both the radiomics signature and independent clinical characteristics. The diagnostic performance of these three models was evaluated in the validation cohort based on their area under the curve (AUC), sensitivity, specificity, and accuracy.ResultsAmong the 491 rectal cancer patients, the prevalence of MSI was 10.39% (51/491). Following ICC analysis, two-sample t test, and LASSO regression, six radiomics features were selected for subsequent analysis. The combined model, which incorporated both the clinical factors and radiomics features achieved an AUC of 0.895 [95% confidence interval (CI), 0.838–0.938] in the validation cohort, and showed better performance in predicting MSI status than the other two models using either clinical factors (P=0.015) or radiomics features (P=0.204) alone.ConclusionsRadiomics features based on preoperative T2-weighted MR imaging (MRI) are associated with the MSI status of rectal cancer. Combinational analysis of clinical factors and radiomics features may improve predictive performance and potentially contribute to noninvasive personalized therapy selection.