可解释的机器学习预测模型用于子宫内膜癌患者的早期识别OA

Objective This study aimed to develop a machine learning(ML)model based on clinical data obtained from noninvasive examinations to predict the risk of endometrial carcinoma(EC)in women of all ages,thereby improving the efficiency of early diagnosis of EC.Methods Clinical data were collected from 344 patients with EC and 344 patients with benign endometrial lesions(including endometrial polyps and endometrial hyperplasia without atypia)treated at Inner Mongolia Maternal and Child Health Care Hospital between January 2020 and March 2025.Sixteen relevant features were identified from the clinical dataset.The dataset was divided into training and validation sets using 10-fold cross-validation.Four models-eXtreme gradient boosting(XGBoost),random forest(RF),logistic regression(LR),and support vector machine(SVM)-were constructed.Model performance was evaluated using the area under the curve(AUC),sensitivity,and specificity to determine the optimal model.In addition,an external test set consisting of 142 patients from other hospitals was used to validate the optimal model and conduct sensitivity analysis.Finally,SHapley Additive exPlanations(SHAP)were applied to interpret the model.Results Among the four models,the XGBoost model demonstrated the best performance across all evaluation metrics and exhibited the strongest overall predictive ability.In the training set,the AUC was 0.96,while in the test set the AUC reached 0.89.SHAP analysis indicated that age,platelet distribution width(PDW),and the triglyceride-glucose index(TyG)were among the most important predictive factors.Conclusion This study developed a predictive model for the early diagnosis of EC and identified several feature indicators associated with the risk of EC.The model may effectively assist clinicians in identifying individuals at high risk and facilitate targeted clinical interventions.