BUS TI-RADS 2.0: an optimized study on a malignant risk stratification system for thyroid nodules based on morphological characteristics of blood flow.

OBJECTIVE: To upgrade the baseline ultrasound thyroid imaging reporting and data system 2.0 (BUS TI-RADS 2.0) by refining nonradiating vascularity subclassification and recalibrating feature weights to enhance diagnostic accuracy and clinical utility. METHODS: The ultrasonic images of 4852 thyroid nodules of 3639 patients in 4 hospitals (2018-2025) were retrospectively analyzed. "Non-radiating vascularity" was subclassified into five types. Multivariable logistic regression identified predictors to construct BUS TI-RADS 2.0. Its performance was compared with the original BUS TI-RADS, American College of Radiology TI-RADS (ACR TI-RADS), and Chinese TIRADS (C-TIRADS), and biopsy-reducing potential was assessed. RESULTS: Multivariate regression confirmed established predictors (e.g., composition, echogenicity, shape, margin, suspected extrathyroidal extension, punctate echogenic foci, vascularity). Diverging from BUS TI-RADS, BUS TI-RADS 2.0 rescored vascularity: circumferential vascularity (-2 points), disorganized vascularity (+1 point), and avascular/punctate/wheel-like vascularity (0 points); "ill-defined margin" was also adjusted to + 1 point. The final score determined risk stratification. BUS TI-RADS 2.0 achieved an AUC of 0.98 in the training set, significantly outperforming BUS TI-RADS (0.97), ACR TI-RADS (0.94), and C-TIRADS (0.93) (all P < 0.001). Internal/external validation confirmed comparable diagnostic value to BUS TI-RADS (AUCs: 0.96 vs. 0.96; 0.94 vs. 0.94) and superiority over others (P < 0.001). The new system markedly reduced false-negative rates (12.98% vs. 18.05%-68.70%), and yielded low overall biopsy recommendations (2.97% vs. 3.53%-34.08%). CONCLUSIONS: With proven diagnostic performance in trained radiologists, BUS TI-RADS 2.0 may effectively inform clinical decision-making. Broader adoption now depends on validating its consistency across unselected cases and varying reader expertise.