Abstract
BACKGROUND: Ultrasound plays a crucial role in the preoperative evaluation of thyroid carcinoma and its extrathyroidal extension (ETE). This study aims to develop and validate a model integrating ultrasound radiomics and clinical factors for predicting ETE. METHODS: In this retrospective study, data from 420 patients who underwent thyroid cancer surgery at the First Affiliated Hospital of Nanchang University were reviewed. Patients were categorized into non-ETE and ETE groups based on postoperative pathology and randomly divided (7:3 ratio) into training and internal validation cohorts. An external test cohort included 70 patients from Huashan Hospital, Fudan University. Radiomics features were extracted from preoperative ultrasound images of the entire tumor region. Feature selection was performed using Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression to construct a radiomics signature. A nomogram was subsequently developed by combining the radiomics score with clinical-ultrasound factors. The model's predictive performance was evaluated using receiver operating characteristic (ROC) curve analysis and decision curve analysis (DCA) across all cohorts. RESULTS: The combined model demonstrated excellent calibration and discrimination, with areas under the curve (AUC) of 0.898, 0.873, and 0.857 in the training, internal validation, and external test cohorts, respectively. DCA confirmed the superior clinical utility of the radiomics nomogram compared to models using clinical-ultrasound factors alone. Subgroup analysis further revealed the model's robust performance across different tumor sizes, ETE degrees, and Hashimoto's thyroiditis status. CONCLUSION: The ultrasound radiomics-based nomogram shows favorable performance and high accuracy in predicting ETE, suggesting its potential as a valuable preoperative auxiliary diagnostic tool.