A predictive diagnostic model using multiparametric MRI for differentiating uterine carcinosarcoma from carcinoma of the uterine corpus

Purpose: To construct a diagnostic model for differentiating carcinosarcoma from carcinoma of the uterus. Materials and methods: Twenty-six patients with carcinosarcomas and 26 with uterine corpus carcinomas constituted a derivation cohort. The following nine MRI features of the tumors were evaluated: inhomogeneity, predominant signal intensity, presence of hyper- and hypointense areas, conspicuity of tumor margin, cervical canal extension on T2WI, presence of hyperintense areas on T1WI, contrast defect area volume percentage, and degree of enhancement. Two predictive models—with and without contrast—were constructed using multivariate logistic regression analysis. Fifteen other patients with carcinosarcomas and 30 patients with carcinomas constituted a validation cohort. The sensitivity and specificity of each model for the validation cohort were calculated. Results: Inhomogeneity, predominant signal intensity on T2WI, and presence of hyperintense areas on T1WI were significant predictors in the unenhanced-MRI-based model. Presence of hyperintensity on T1WI, contrast defect area volume percentage, and degree of enhancement were significant predictors in the enhanced-MRI-based model. The sensitivity/specificity of unenhanced MRI were 87/73 and 87/70% according to reviewer 1 and 2, respectively. The sensitivity/specificity of the enhanced-MRI-based model were 87/70% according to both reviewers. Conclusions: Our diagnostic models can differentiate carcinosarcoma from carcinoma of the uterus with high sensitivity and moderate specificity.