Using Machine Learning to Predict the Prognosis in Endometrial Cancer Patients Undergoing Fertility-Sparing Treatment

Endometrial cancer is a significant gynecological malignancy with rising incidence among women of reproductive age, necessitating effective fertility-sparing treatments. Machine learning offers potential in enhancing prognostic assessments through radiomics.

To develop and validate a machine learning-based model integrating radiomics and clinical features to predict the prognosis of fertility-sparing treatments in endometrial cancer patients.

This retrospective study included 102 endometrial cancer patients who received fertility-sparing treatment at West China Second University Hospital from November 2017 to December 2023. Patients were randomly divided into training (n = 81) and testing (n = 21) cohorts. The primary outcome was 6 month treatment response evaluated by endometrial sampling. MRI-based radiomic features and clinical data were analyzed using logistic regression and other machine learning algorithms. A combined model was constructed by incorporating radiomics and clinical features. Model performances were assessed by the area under the receiver operating characteristic curve (AUC) analysis, and decision curve analysis (DCA) was used to estimate the models' clinical values.

Among 102 patients, 60 (58.8%) achieved complete remission (CR), whereas 42 (41.2%) did not (non-CR). In total, 1197 radiomic features were extracted from MRI images. Thirteen radiomic features were deemed valuable by dimensionality reduction and selection. Among radiomic models, the logistic regression model was the most effective, showing high stability and accuracy, with AUCs of 0.875 in the training cohort and 0.852 in the test cohort. Ultimate clinical features selected by univariate and multivariable logistic regression constructed a clinical model. The combined model demonstrated superior performance with an AUC of 0.941 in the training cohort and 0.907 in the testing cohort. Combined model DCA revealed optimal clinical efficacy.

The integrated model effectively predicts fertility preservation outcomes, offering a reliable noninvasive tool for clinicians. This approach may enhance personalized treatment strategies for young endometrial cancer patients.