Machine learning models for prediction of NPVR ≥80% with HIFU ablation for uterine fibroids

Currently high-intensity focused ultrasound (HIFU) is widely used to treat uterine fibroids (UFs). The aim of this study is to develop a machine learning model that can accurately predict the efficacy of HIFU ablation for UFs, assisting the preoperative selection of suitable patients with UFs.

This study collected data from 1,000 patients with UFs who underwent ultrasound-guided high-intensity focused ultrasound. The least absolute shrinkage and selection operator (LASSO) regression was used for multidimensional feature screening. Five machine learning algorithms such as logistic regression, random forest, extreme gradient boosting (XGBoost), artificial neural network, and gradient boosting decision tree were utilized to predict ablation efficacy. The efficacy was quantified by the non-perfused volume ratio (NPVR), which was classified into two categories: NPVR <80% and NPVR ≥80%.

The XGBoost model proved to be the most effective, showing the highest AUC of 0.692 (95% CI: 0.622-0.762) in the testing data set. The four key predictors were T2 weighted image, the distance from ventral side of UFs to skin, platelet count, and contrast-enhanced T1 weighted image.

The machine learning prediction model in this study showed significant potential for accurately predicting the preoperative efficacy of HIFU ablation for UFs. These insights were important for clinicians in the preoperative assessment and selection of patients, which could enhance the precision of treatment planning.