Abstract
OBJECTIVE: This study aimed to develop a nomogram integrating radiomics features derived from contrast-enhanced ultrasound (CEUS) and B-mode ultrasound (B-US) with clinical features to improve preoperative differentiation between follicular thyroid carcinoma (FTC) and follicular thyroid adenoma (FTA). Accurate preoperative diagnosis is critical for guiding appropriate treatment strategies and reducing unnecessary interventions. METHODS: We retrospectively included 201 patients with histopathologically confirmed FTC (n = 133) or FTA (n = 68). Radiomics features were extracted from B-US and CEUS images, followed by feature selection and machine-learning model development. Five models were evaluated, and the one with the highest area under the curve (AUC) was used to construct a radiomics signature. A Clinical Risk model was developed using statistically significant clinical features, which outperformed the conventional Chinese Thyroid Imaging Reporting and Data System (C-TIRADS) in both training and test groups. The radiomics signature and Clinical Risk model were integrated into a nomogram, whose diagnostic performance, calibration and clinical utility were assessed. RESULTS: The Clinical Risk model achieved superior diagnostic performance compared to the C-TIRADS model, with AUCs of 0.802 vs. 0.719 in the training group and 0.745 vs. 0.703 in the test group. The nomogram further improved diagnostic efficacy, with AUCs of 0.867 (95% CI, 0.800-0.933) in the training group and 0.833 (95% CI, 0.729-0.937) in the test group. It also demonstrated excellent calibration. Decision curve analysis (DCA) also indicated that the nomogram showed good clinical utility. CONCLUSION: By combining CEUS and B-US radiomics features with clinical data, we developed a robust nomogram for distinguishing FTC from FTA. The model demonstrated superior diagnostic performance compared to existing methods and holds promise for enhancing clinical decision-making in thyroid nodule management. CLINICAL TRIAL NUMBER: Not applicable.