Predicting the regrowth of residual uterine fibroids after high-intensity focused ultrasound treatment: an interpretable magnetic resonance imaging radiomics model.

Abstract

Background: The evaluation of residual uterine fibroids (RFs) after magnetic resonance imaging (MRI)-based radiomics is complex, making it challenging to accurately predict and interpret the regrowth of RFs following high-intensity focused ultrasound (HIFU) treatment. Therefore, the aim of this research was to establish a robust multiparametric radiomics model which functions to predict the regrowth of RFs following HIFU treatment. Moreover, SHapley Additive exPlanations (SHAP) was adopted to clarify the internal prediction process of the model.

Methods: In this retrospective investigation, 116 patients diagnosed with uterine fibroids who underwent HIFU treatment were enrolled, and underwent follow-up imaging approximately one-year post-treatment. Patients were categorized into RF regrowth and non-regrowth groups based on the occurrence of residual fibroid regrowth 1 year after treatment. The cohort was divided into a training set (N=92) and a test set (N=24). A total of 218 radiomic features were acquired from T2-weighted imaging (T2WI) and contrast-enhanced T1-weighted imaging (CE-T1WI) scans. Subsequent to the implementation of preprocessing and feature selection steps, logistic regression (LR) models were developed using radiomic features from T2WI and CE-T1WI, as well as a feature-level fusion of both. Finally, the SHAP approach was applied to interpret the underlying predictive mechanisms.

Results: The LR models achieved areas under the curve (AUCs) of 0.926 [95% confidence interval (CI): 0.817-1.000] for the T2WI model, 0.879 (95% CI: 0.731-1.000) for the CE-T1WI model, and 0.946 (95% CI: 0.897-0.995) for the fusion model. The SHAP technology was employed to facilitate clinicians' comprehension of the influence exerted by radiomic features on the model's predictions from both global and individual perspectives.

Conclusions: The multiparametric radiomics model demonstrated robustness in predicting the regrowth of RFs post-HIFU treatment. Radiomic features may serve as potential biomarkers for preoperative evaluation for HIFU treatment and enhance the mechanism of RF regrowth after HIFU.