Multimodal ultrasound to predict central cervical lymph node metastasis in clinically lymph node-negative differentiated thyroid carcinomas: a two-center retrospective study.

[BACKGROUND] The incidence of thyroid cancer has been on the rise in recent decades. The preoperative diagnosis of central cervical lymph node metastasis (CCLNM) has been found to be associated with thyroid cancer recurrence and patient prognosis, as well as impacting surgical technique and scope decision. Therefore, it is imperative to have a precise preoperative assessment of CCLNM. Therefore, we developed a reliable predictive model which incorporates preoperative clinical data and multimodal ultrasound (US) features to accurately predict CCLNM in clinically lymph node-negative (cN0) T1/T2 differentiated thyroid carcinoma (DTC) patients.

[METHODS] Patients diagnosed with DTC confirmed through pathological results between July 2023 and February 2025 were included in the retrospective study. The 613 patients were divided into the training cohort (n=357, from Shanghai General Hospital), internal validation cohort (n=153, from Shanghai General Hospital), and external validation cohort (n=103, from Xuzhou Centeral Hospital). Univariate and multivariable logistic regression analyses were used for assessing multimodal US variables associated with CCLNM, and a predictive model was constructed from the multivariable analysis. Predictive performance was assessed with the receiving operating characteristic (ROC) curve and was further evaluated in the internal and external validation datasets.

[RESULTS] Patient age, tumor size, echogenicity, discontinuous capsule enhancement (DCE), and strain rate ratio (SRR) were found to be independently associated with CCLNM. The multimodal US model showed optimal diagnostic performance [area under the receiver operating characteristic curve (AUC), 0.84; 95% confidence interval (CI): 0.80, 0.88] in predicting CCLNM, and demonstrated good discrimination in both the internal (AUC, 0.81; 95% CI: 0.75, 0.88) and external validation cohorts (AUC, 0.80; 95% CI: 071, 0.87).

[CONCLUSIONS] Our proposed predictive model based on preoperative clinical and multimodal US parameters carries the potential to predict CCLNM in cN0 DTCs, and may assist in minimizing the risk of unnecessary prophylactic central lymph node dissection (PCLND) among such patients.