To evaluate the role of boosting based ML algorithms for predicting the treatment outcome of HIFU ablation with an NPV ratio of at least 90%.

We, in this study, used multiparametric (mp) MRI features including anatomical characteristics of patients and tissue characteristics of lesion outlining its cellularity, diffusivity, and vascularity. Five ML classifiers —including Adaboost classifier, Gradient boosting (GBM) classifier, XGBOOST classifier, CatBoost classifier, LightGBM Classifier, HistGradientBoosting classifier and ensemble model that consist of the combination of these classifiers were used. For the evaluation of the classifier performance, cross validation score was utilized.

The whole data set of 73 patients was split into 4 groups by performing the 4-fold cross validation to evaluate generalizability and accuracy of the model.  The best classifier performance was achieved using ensemble model whose cross-validation score was 0.96. XGBOOST classifier showed the secondary predictive performance with cross validation score 0.95, followed by HistGradientBoosting, Adaboost, CatBoost, LightGBM and GBM classifiers (cross validation score: 0.93, 0.92, 0.92, 0.90 and 0.89), respectively.

This preliminary study indicates that ML algorithms should be considered in assisting physicians to fully evaluate the outcome of the HIFU therapy.