Ultrasound-based radiomics for prediction of lateral neck lymph node metastasis in patients with medullary thyroid cancer.

Background: Routine lateral neck lymph node dissection in patients with medullary thyroid cancer (MTC) remains controversial. Therefore, we sought to develop a model to predict the risk of lateral neck lymph node metastasis (LLNM) in patients with MTC. Methods: This retrospective study analyzed the clinical, ultrasound, and radiomics features of patients who underwent surgery and were diagnosed with MTC in our hospital from 2014 to 2023. The main outcome was LLNM, for which three models were developed: conventional model based on clinical and ultrasound features, radiomics score (rad score) model, and nomogram model that combined clinical, ultrasound, and radiomics features. Internal validation was performed using the bootstrap method, and the results were compared. The performance was calculated using area under the receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis. Results: There were 89 patients included, of whom 35 patients had LLNM. No significant associations were found between gender or age and LLNM (P>0.05). The increase in calcitonin (>539.7 pg/mL) was associated with the LLNM (P=0.07). Two preoperative ultrasound findings, subcapsular location and suspicious lateral lymph nodes on ultrasound, were associated with the LLNM (P<0.05). Five radiomics features were selected to develop the rad score model. The nomogram model had a higher AUC compared with the conventional model and rad score model (P=0.008, 0.01, respectively). Conclusions: This study developed a nomogram model that integrated clinical, ultrasound, and radiomics features that could be used to predict LLNM in patients with MTC.