Predicting Malignancy in Solid Adnexal Masses: An Externally Validated Machine Learning Model Integrating Conventional and Contrast-Enhanced Ultrasound

Background Traditional diagnostic models, or contrast-enhanced ultrasound-related models, are generally developed for broad applications across different adnexal masses. These broad models lack the precision needed to describe specific subtypes, such as solid ones. Objective This study aimed to develop a machine learning model using conventional and contrast-enhanced ultrasound features to stratify malignancy in solid adnexal masses and to evaluate its performance relative to updated O-RADS on an independent external test set. Methods A total of 277 solid adnexal masses were analysed in the development set. Missing data were addressed through multiple imputation, generating 20 datasets (m = 20). Feature selection was conducted using bootstrap-enhanced least absolute shrinkage and selection operator (LASSO) regression to retain the nine most stable predictors. Model performance and discrimination were assessed on three datasets. In the external test set, our model showed improved discrimination compared to updated O-RADS classifications, as determined by the DeLong test. Decision curve analysis (DCA) further demonstrated its greater clinical applicability. Results In internal validation, the model showed robust discrimination (AUC: 0.926; 95% CI: 0.861–0.992) and good calibration (Brier score: 0.115). These results were further confirmed in the external test set (AUC: 0.949, 95% CI: 0.906–0.992) and with good calibration (Brier score: 0.101). The novel model outperformed the updated O-RADS categories in the external test set for O-RADS = 4 (AUC: 0.949 vs 0.656, DeLong test, p  < 0.001) and O-RADS = 5 (AUC: 0.949 vs 0.593, DeLong test, p  < 0.001). Decision curve analysis indicated that the model was suitable across a range of clinical trial thresholds.providing flexible thresholds, and building a visualised nomogram, Conclusions The new model, integrating conventional ultrasound and CEUS features, can enhance diagnostic accuracy for solid adnexal masses, surpassing the updated O-RADS categorisation and providing increased practical value for clinical assessments.