Postmenopausal endometrial non-benign lesion risk classification through a clinical parameter-based machine learning model

This study aimed to develop and evaluate a machine learning model utilizing non-invasive clinical parameters for the classification of endometrial non-benign lesions, specifically atypical hyperplasia (AH) and endometrioid carcinoma (EC), in postmenopausal women. Our study collected clinical parameters from a cohort of 999 patients with postmenopausal endometrial lesions and conducted preprocessing to identify 57 relevant characteristics from these irregular clinical data. To predict the presence of postmenopausal endometrial non-benign lesions, including atypical hyperplasia and endometrial cancer, we employed various models such as eXtreme Gradient Boosting (XGBoost), Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), Back Propagation Neural Network (BPNN), as well as two ensemble models. Additionally, a test set was performed on an independent dataset consisting of 152 patients. The performance evaluation of all models was based on metrics including the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, precision, and F1 score. The RF model demonstrated superior recognition capabilities for patients with non-benign lesions compared to other models. In the test set, it attained a sensitivity of 88.1% and an AUC of 0.93, surpassing all alternative models evaluated in this study. Furthermore, we have integrated this model into our hospital's Clinical Decision Support System (CDSS) and implemented it within the outpatient electronic medical record system to continuously validate and optimize its performance. We have trained a model and deployed a system with high discriminatory power that may provide a novel approach to identify patients at higher risk of postmenopausal endometrial non-benign lesions who may benefit from more tailored screening and clinical intervention.