Multimodal MRI radiomics for predicting HIFU ablation efficacy in uterine fibroids: a machine learning study

To explore the predictive value of machine learning-based multimodal MRI radiomics combined with clinical features in the efficacy of high-intensity focused ultrasound (HIFU) ablation of uterine fibroids.

This study included 390 patients with uterine fibroids who underwent HIFU ablation. Patients were stratified into high and low ablation groups based on an 80% non-perfused volume ratio (NPVR) and randomly divided into training (70%) and test (30%) sets. Radiomics features were extracted from T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), and diffusion-weighted imaging (DWI). The most predictive features were selected via Recursive Feature Elimination (RFE) and the Least Absolute Shrinkage and Selection Operator (LASSO), and combined with clinical characteristics. Logistic Regression (LR), Random Forest (RF) and eXtreme Gradient Boosting (XGBoost), were constructed to predict ablation efficacy, with performance assessed using the area under the receiver operating characteristic curve (AUC).

The results indicated that age, uterine fibroid location, and T2WI signal intensity were independent predictive factors (p < 0.05). The multimodal-clinical fusion XGBoost model exhibited the optimal performance. In the test set, this model achieved an AUC of 0.894, with an accuracy of 82.1%, sensitivity of 88.9%, and specificity of 74.1%.The calibration curve and decision curve analysis (DCA) confirmed that the predicted probabilities of the model were highly consistent with the actual risks, and stable clinical net benefits were achieved.

The XGBoost model based on multimodal MRI and clinical features may serve as a reference for predicting HIFU ablation efficacy and optimizing treatment strategies.