A study aimed to develop and validate a CT-based radiomics model to differentiate benign and malignant thyroid nodules. It involved 378 patients with 408 resected nodules from two centers, divided into training, internal validation, and external validation sets. Various models were created using clinical, morphologic CT features, and radiomic features from non-contrast, arterial, and venous phase CTs. The combined model, which included patient age, three morphologic features, and 28 radiomic features, demonstrated the highest performance. In the external validation set, it achieved an AUC of 0.923, with a sensitivity of 84.0%, specificity of 94.1%, and accuracy of 87.0%. It outperformed other models significantly (all p < .05) and showed the best calibration and highest net benefit in decision-curve analysis. This combined model may aid in managing thyroid nodules detected on CT.
