A predictive model combining contrast-enhanced ultrasound and Shamblin classification for risk stratification of internal carotid artery resection in carotid body tumor surgery

Background: Carotid body tumors (CBTs) are rare paragangliomas at the carotid bifurcation. Surgical resection is the primary treatment; however, internal carotid artery (ICA) resection may be required and carries high risks of cerebral ischemia and complex reconstruction. The conventional Shamblin classification, based on arterial encasement, does not reliably predict ICA resection. This study aimed to evaluate whether the tumor-ICA interface on contrast-enhanced ultrasound (CEUS) can serve as a predictor of ICA resection, and to develop a combined predictive model for risk stratification.

Methods: This prospective study included 54 patients with 59 CBT lesions who underwent preoperative CEUS and subsequent surgery at a tertiary referral center between December 2022 and March 2025. The CEUS interface was qualitatively assessed as present or absent. The primary outcome was intraoperative ICA management (preservation vs. resection). Univariate analyses were performed, followed by multivariate analysis using Firth’s penalized‑likelihood regression. Predictive models based on each independent predictor and their combination were constructed, and their performance was compared using the area under the curve (AUC) of the receiver operating characteristic (ROC). A risk-stratification system was then established based on the combined model.

Results: ICA resection was required in 12 of the 59 lesions (20.3%). The multivariate analysis identified the CEUS interface [odds ratio (OR) 40.46, 95% confidence interval (CI): 4.55–1190.09, P<0.001] and the Shamblin grade (OR 11.92, 95% CI: 1.02–314.23, P=0.049) as independent predictors of ICA resection in CBT surgery. The combined model achieved an AUC of 0.949 (95% CI: 0.860–1.000), compared with 0.906 (95% CI: 0.794–1.000) for the CEUS interface alone and 0.843 (95% CI: 0.710–0.976) for the Shamblin grade alone. The model stratified patients into four distinct risk tiers, with the predicted probability of ICA resection ranging from 3.1% (CEUS interface present and Shamblin grade I/II) to 96.3% (CEUS interface absent and Shamblin grade III).

Conclusions: The CEUS interface is a strong preoperative predictor of ICA resection in CBT surgery. When combined with the Shamblin grade, it provides a promising predictive model for risk stratification that may aid in surgical planning. However, external validation is needed before clinical implementation.