Research on predicting vascular infiltration of endometrial cancer based:on multi-parameter MRI radiomics combined with clinical and pathological features

LAI Jiaxin; LI Yuchen; LIU Wei; YAN Rui

doi:10.12122/j.issn.1674-4500.2025.07.03

Abstract：Objective To develop a fusion model that integrates multiparametric magnetic resonance imaging(mpMRI)-based radiomics features with clinical and pathological variables for predicting lymph-vascular space invasion(LVSI)in patients with endometrial cancer.Methods This retrospective study included 96 patients with pathologically confirmed endometrial cancer treated at Northwest Women's and Children's Hospital from January 2015 to June 2024.Axial T2-weighted imaging(T2WI)and diffusion-weighted imaging(DWI)sequences were used to manually delineate both tumor lesions and corresponding uterine body regions.radiomics features were extracted from the delineated regions.clinical and pathological variables were screened using univariate analysis and significant predictors were integrated with imaging features to construct a fusion model.model performance was evaluated using leave-one-out cross-validation.Results The AUC for DWI-based radiomics models reached 0.84 for tumor lesions and 0.87 for the uterine body,while the T2WI-based models yielded AUCs of 0.82 and 0.84,respectively.multivariate logistic regression identified age,CA199 and Ki67 expression as independent predictors of LVSI(P<0.05),with the combined clinical-pathological model achieving an AUC of 0.834.The final fusion model,incorporating both radiomics and clinical-pathological features,achieved an AUC of 0.920,demonstrating superior predictive performance compared to single-modality models.Conclusion The integration of mpMRI-derived radiomics with key clinical and pathological factors significantly enhances the predictive accuracy for LVSI in endometrial cancer.this fusion approach may provide valuable support for accurate preoperative staging and the development of individualized treatment strategies.