Machine learning-based predictive analysis of energy efficiency factors necessary for the HIFU treatment of adenomyosis.

Purpose: This study aimed to develop a joint model combining T2-weighted imaging (T2WI) suppressed fat radiomics, and clinical parameters to predict the energy efficiency factor (EEF) required for high-intensity focused ultrasound (HIFU) ablation in patients with adenomyosis.

Materials And Methods: This retrospective study included 169 adenomyosis patients who underwent HIFU ablation between September 2021 and May 2024. EEF values were calculated based on T2WI fat suppression (T2WI-FS) sequences, and radiomics features were extracted. Predictive features were selected using minimum redundancy maximum relevance (MRMR) and least absolute shrinkage and selection operator (LASSO) methods, and two joint-based on decision tree and random forest algorithms-models were developed for EEF prediction.

Results: The decision tree model achieved a mean absolute error (MAE) of 8.095 on the test set, while the random forest model exhibited an MAE of 8.231. The Wilcoxon rank-sum test for the test set revealed that the discrepancy in predictive performance between the two models was statistically significant ( < 0.05). The correlation coefficients were 0.768 and 0.777, and the coefficients of the two models in the test set were 0.559 and 0.549, respectively.

Conclusion: The joint model integrating T2WI radiomics and clinical data effectively predicted EEF values for HIFU ablation in adenomyosis. This approach provides a foundation for optimizing HIFU dosing strategies and enhancing treatment safety and efficacy.