Purpose: The aim of this study was to evaluate the value of different multiparametric MRI-based radiomics models in differentiating TP53-mutant from TP53-wild-type endometrial carcinoma (EC).
Materials and Methods: A total of 802 endometrial carcinoma (EC) patients were included and divided into a training group (n = 642) and a test group (n = 180). Clinical predictors were screened using logistic regression analysis. Radiomics features were extracted and selected from multiparametric magnetic resonance (MR) images. Following data dimensionality reduction and feature selection, eleven machine learning algorithms were applied to construct models to identify the optimal radiomics model for differential diagnosis. The diagnostic performance of the nomogram was evaluated using the receiver operating characteristic (ROC) curve. The clinical benefit of the nomogram was assessed through decision curve analysis (DCA).
Results: In the T2-sagittal (T2-sag) sequence, 834 radiomics features were extracted from each image. Among 11 machine learning algorithms, the ExtraTrees model demonstrated optimal radiomics performance in both the training and validation sets, achieving the highest mean AUC (0.727) and accuracy (0.790). In the T2-coronal (T2-cor) sequence, 2 ineligible patients were excluded; 1,198 features were extracted per image. The RandomForest model outperformed others, yielding the highest mean AUC (0.731) and accuracy (0.793). For the T2-axial (T2-oax) sequence, 1 patients were excluded; 1,198 features were extracted. The ExtraTrees model again showed superior performance with a mean AUC( 0.745) and accuracy (0.702). Finally, in the ADC sequence, 26 patients were excluded; 1,198 features were extracted. The ExtraTrees model exhibited the best performance, attaining a mean AUC (0.761) and accuracy (0.712).
Conclusion: Multiparametric MRI-based radiomics models can non-invasively differentiate TP53-mutant from TP53-wild-type endometrial carcinoma (EC), providing enhanced auxiliary diagnostic value for clinical applications.