Multimodal ultrasound radiomics for predicting cervical lymph node metastasis in solitary papillary thyroid carcinoma: a multicenter study

The accurate preoperative assessment of cervical lymph node metastasis (CLNM) in patients with papillary thyroid carcinoma (PTC) is crucial for determining the optimal surgical strategy. However, traditional ultrasound methods have limitations in sensitivity and specificity. This study aims to develop and validate a predictive nomogram for CLNM using multimodal ultrasound radiomics. This retrospective multicenter study included 309 PTC patients. The samples were divided into training set, validation set, and external test set. Radiomics features were extracted from strain elastography (SE), sound touch elastography (STE), and B-mode ultrasound (BMUS) images. Five machine learning models were constructed based on the selected radiomics features, alongside the development of a clinical prediction model based on clinical ultrasound characteristics. A combined clinical-radiomics nomogram was developed by integrating the optimal ML model with clinical predictors. The performance of the models was analyzed based on accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC), as well as the decision curve analysis (DCA). The combined clinical-radiomics nomogram outperformed both the ML models and the clinical model in diagnosing lymph node metastasis in PTC. In the training set, the AUC, accuracy, sensitivity, and specificity were 0.877 (95% CI 0.818–0.927), 0.797, 0.790, and 0.802, respectively. In the validation set, these values were 0.928 (95% CI 0.855–0.983), 0.882, 0.926, and 0.854, respectively. In the external test set, the values were 0.844 (95% CI 0.751–0.925), 0.795, 0.875, and 0.745, respectively. The DCA also confirmed the clinical usefulness of the combined clinical-radiomics nomogram. Integrating multimodal ultrasound radiomics with routinely available clinical and elastographic predictors may improve preoperative CLNM risk stratification in patients with solitary PTC. The resulting nomogram may support individualized cervical lymph-node assessment and surgical planning; however, prospective multicenter validation under standardized ultrasound acquisition protocols is required before routine clinical implementation.