A MRI-based nomogram to preoperatively predict histologic grades in stage I endometrial carcinoma

Background: This study aims to develop and externally validate a multi-sequence MRI-based radiomics nomogram for preoperatively differentiating low-grade and high-grade tumors in FIGO stage I endometrial carcinoma (EMC). Methods: A primary cohort was established with a total of 100 patients enrolled from our hospital between Jan. 2017 and Apr. 2021. A consecutively enrolled internal validation cohort (n=41) and an external validation cohort (n=50) were used to test the models. Radiomics features were extracted from T1-weighted contrast-enhanced (T1-CE) and T2-weighted (T2W) MRI and selected with the least absolute shrinkage and selection operator (LASSO) to build the radiomics signature (RS). Clinical factors were analyzed with Mann-Whitney U test and Chi-square test. A radiomics nomogram model was constructed incorporating the RS and the most discriminative clinical parameter. Performance of the RS, clinical parameter and nomogram were validated with receiver operating characteristic (ROC), calibration and decision curve analysis (DCA). Results: : The multi-sequence MRI-based RS was built integrating 3 selected features, all from T1-CE MRI. The deep myometrial invasion was considered as the most important clinical parameter ( P <0.05). The nomogram model incorporates RS and deep myometrial invasion yielded the best discriminative performance with AUCs of 0.845 (95% confidence interval [CI] 0.759-0.910, SEN=0.900, SPE =0.650), 0.756 (95% CI 0.597-0.876, SEN=0.818, SPE =0.632) and 0.779 (95% CI 0.639-0.884, SEN=0.800, SPE =0.720) in the primary, internal validation and external validation cohorts, respectively. Calibration curves and DCA suggested good potential of our nomogram in clinical uses. Conclusions: : The developed radiomics nomogram can be used as a potential non-invasive tool for preoperatively differentiating low- and high-grade tumors in stage I EMC patients.