Preoperative prediction of lateral cervical lymph node metastasis in papillary thyroid carcinoma using conventional ultrasound, sonazoid contrast enhanced ultrasound, fine needle aspiration thyroglobulin and radiomics

Abstract

Accurate preoperative identification of lateral cervical lymph node metastasis (LLNM) is crucial for surgical decision-making and prognostic assessment in patients with papillary thyroid carcinoma (PTC). Because prophylactic lateral neck dissection is not routinely recommended owing to the risks of surgical complications and cosmetic impairment, reliable preoperative confirmation of LLNM remains an important clinical challenge. This study aimed to comprehensively evaluate the diagnostic performance of conventional ultrasound, Sonazoid contrast-enhanced ultrasound (CEUS), fine- needle aspiration thyroglobulin (FNA-Tg), and ultrasound-based radiomics for the preoperative prediction of LLNM in PTC. The final analytical cohort comprised 178 patients. Of these, 44 had undergone the relevant examinations as part of routine clinical care between 1 February and 13 June 2024 and were retrospectively included only after local ethical approval was obtained on 14 June 2024. A further 134 patients were enrolled from July 2024 to November 2025 under the approved study protocol. Each patient contributed one lateral cervical lymph node showing the most suspicious or pronounced imaging abnormalities for analysis. All selected lymph nodes underwent conventional ultrasound, Sonazoid CEUS assessment during the arterial, venous, and post-vascular phases, and FNA-Tg measurement. Qualitative and quantitative diagnostic approaches were compared using receiver operating characteristic analysis and paired DeLong tests. Multimodal combinations of conventional ultrasound, Sonazoid CEUS, and FNA-Tg were also evaluated. Radiomics features were extracted from grayscale ultrasound, colour Doppler flow imaging, and images from the three CEUS phases. Eleven machine-learning algorithms were assessed using the area under the receiver operating characteristic curve, decision curve analysis, calibration curves, and sample-level prediction scores. Quantitative approaches generally showed higher diagnostic performance than the corresponding qualitative assessments for conventional ultrasound, multiphase Sonazoid CEUS, and FNA-Tg. Sonazoid CEUS achieved an AUC of 0.979 and significantly outperformed conventional ultrasound (p = 0.0057). Among the multimodal strategies, the combination of conventional ultrasound, Sonazoid CEUS, and FNA-Tg yielded the highest observed AUC of 0.993, although it was not statistically superior to the high-performing two-modality combinations. Radiomics performance varied across imaging modalities. The multilayer perceptron model based on post-vascular-phase CEUS achieved the highest observed test-cohort AUC of 0.882 among the radiomics models evaluated. These findings support a risk- adapted diagnostic approach in which conventional ultrasound provides initial anatomical assessment, Sonazoid CEUS adds functional information, FNA-Tg provides biochemical confirmation, and radiomics offers additional quantitative risk stratification. External multicentre validation is required before the radiomics models or integrated pathway can be adopted routinely.