Intratumoral and peritumoral multiparametric MRI-based radiomics nomogram for preoperative risk stratification in patients with endometrial cancer.

Introduction: Achieving accurate preoperative risk stratification for endometrial cancer (EC) is challenging due to the need for histopathology to obtain the necessary parameters. This study aimed to establish and validate a multiparametric magnetic resonance imaging (MRI) radiomics nomogram that incorporates the peritumoral region for preoperative risk stratification in EC patients. Methods: Three-hundred seventy-four women with histologically confirmed EC were divided into training (1.5-T MRI, n=163), test (1.5-T MRI, n=70), and independent validation (3.0-T MRI, n=141) cohorts. As per the guidelines of the European Society of Medical Oncology, patients were categorized into four risk groups: low, intermediate, high-intermediate, and high. Binary classification models were subsequently constructed to distinguish between low- and non-low-risk individuals. Radiomic features were extracted from intra- and peritumoral regions via T2-weighted imaging (T2WI) and apparent diffusion coefficient (ADC) maps. Feature selection was carried out via univariate analysis, least absolute shrinkage and selection operator (LASSO) regression, and multivariate logistic regression. A radiomic model (radscore) was established using the selected features. A nomogram combining the radscore and most predictive clinical parameters was developed. Decision curve analysis (DCA) and the net reclassification index (NRI) were used to assess the clinical benefit of the nomogram. Results: Nine radiomic features were selected from intra- and peritumoral regions on ADC maps and T2W images. The nomogram combining the radscore, age, maximum anteroposterior tumor diameter on sagittal T2WI (APsag), and the tumor area ratio (TAR), achieved the highest area under the curve (AUC) values across all cohorts (training: 0.949, test: 0.947, independent validation: 0.909). The nomogram demonstrated superior performance compared to the radscore (AUCtraining = 0.929, AUCtest = 0.917, and AUCindependent validation = 0.813) alone and the clinical model (AUCtraining = 0.855, AUCtest = 0.845, and AUCindependent validation = 0.842). DCA and the NRI demonstrated that the nomogram achieved greater diagnostic performance and net clinical benefits than did the radscore alone. Conclusion: The developed MRI radiomics nomogram achieved high diagnostic performance in classifying low- and non-low-risk EC preoperatively. This tool could provide valuable support for therapeutic decision-making and demonstrates robustness across various field strength data, increasing its potential for widespread clinical application.