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

Area of Science:

• Gynecologic Oncology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Postmenopausal endometrial lesions require accurate classification to guide clinical management.
• Distinguishing between benign conditions, atypical hyperplasia (AH), and endometrioid carcinoma (EC) is crucial.

Purpose of the Study:

• To develop and evaluate a machine learning model for classifying non-benign endometrial lesions in postmenopausal women.
• Utilize non-invasive clinical parameters for predicting atypical hyperplasia (AH) and endometrioid carcinoma (EC).

Main Methods:

• Collected clinical data from 999 postmenopausal patients, identifying 57 relevant features.
• Compared various machine learning models including Random Forest (RF), XGBoost, and logistic regression.
• Validated model performance on an independent dataset of 152 patients using AUC, sensitivity, and specificity.

Main Results:

• The Random Forest (RF) model achieved the highest performance, with 88.1% sensitivity and 0.93 AUC on the test set.
• RF demonstrated superior recognition capabilities for non-benign endometrial lesions compared to other evaluated models.
• The model was integrated into a Clinical Decision Support System (CDSS) for ongoing validation.

Conclusions:

• A machine learning model, particularly RF, shows high discriminatory power for identifying at-risk postmenopausal patients.
• This approach offers a novel strategy for risk stratification, potentially improving screening and clinical intervention.
• The deployed system facilitates continuous performance validation and optimization.