Ultrasound Radiomics with Random Forest for Bethesda IV Nodule Risk Stratification

In a retrospective study of 69 Bethesda IV thyroid nodules, ultrasound radiomics with a Random Forest model achieved an AUC of 0.735 (95% CI 0.513-0.956) on a held-out test set, but wide CIs highlight need for larger validation.

• Random Forest (50 estimators, max depth 10) outperformed KNN and XGBoost, with sensitivity 0.714 and specificity 0.571 on the held-out test set.
• XGBoost failed to generalize (test AUC 0.378), and leave-one-out cross-validation on the training set yielded AUCs ranging 0.397–0.588, emphasizing the need for larger cohorts.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.