A prognostic model for thermal ablation of benign thyroid nodules based on interpretable machine learning

Area of Science:

• Endocrinology
• Medical Imaging
• Artificial Intelligence

Background:

• Increasing detection of benign thyroid nodules necessitates effective symptom management.
• Ultrasound-guided thermal ablation offers volume reduction for symptomatic nodules.
• Predicting ablation efficacy is challenging due to individual variability in absorption.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the efficacy of thermal ablation in benign thyroid nodules.
• To identify key clinical and ultrasonic characteristics influencing nodule volume reduction ratio (VRR).

Main Methods:

• Prospective study involving 518 patients undergoing ultrasound-guided thermal ablation.
• Construction and evaluation of six machine learning models (logistic regression, SVM, decision tree, random forest, XGBoost, LGBM).
• Application of SHapley Additive exPlanations (SHAP) for model interpretability and feature importance analysis.

Main Results:

• The eXtreme Gradient Boosting (XGBoost) model achieved high predictive performance (accuracy 78.9%, AUC 0.86).
• Top predictors for VRR included: proportion of solid components < 20%, initial nodule volume, blood flow score, peripheral blood flow pattern, and proportion of solid components 50-80% (in the satisfactory group).
• 518 nodules were analyzed, with 356 in the satisfactory VRR group (≥70% at 1 year) and 162 in the unsatisfactory group.

Conclusions:

• Interpretable machine learning models can effectively predict VRR after thermal ablation for benign thyroid nodules.
• These models provide valuable insights for preoperative treatment decisions.
• Understanding key influencing factors aids in personalized patient management.