Feed

BackgroundPreoperative differentiation of thyroid follicular tumors remains challenging because cytology has a limited ability to assess capsular and vascular invasion. This study aimed to develop a nomogram integrating multimodal ultrasound and clinical features to predict the risk of malignancy in follicular thyroid tumors.MethodsA total of 483 patients with pathologically confirmed follicular thyroid tumors were retrospectively enrolled. Follicular thyroid adenoma was defined as the benign group, whereas follicular tumor of uncertain malignant potential and follicular thyroid carcinoma were combined into the high-risk group. Clinical data and multimodal ultrasound features were collected. Independent predictors were identified using univariate and multivariate logistic regression analyses, and a nomogram was constructed. Model performance was evaluated using the area under the receiver operating characteristic curve, calibration curve, Hosmer-Lemeshow test, and decision curve analysis.ResultsSex, lesion multiplicity, preoperative thyroglobulin, halo characteristics, "nodule-in-nodule" appearance, and trabecular structure were identified as independent predictors of malignancy (all p < 0.05). The nomogram demonstrated excellent discrimination, with an area under the receiver operating characteristic curve of 0.94 (95% confidence interval: 0.90-0.97). Internal validation using 1000 bootstrap resamples yielded a mean area under the receiver operating characteristic curve of 0.928 (95% confidence interval: 0.915-0.934). Calibration analysis showed good agreement between predicted and observed outcomes (Hosmer-Lemeshow test, p = 0.578). Decision curve analysis indicated a significant net clinical benefit across a threshold probability range of 15%-85%.ConclusionIn this exploratory retrospective study, the proposed nomogram showed preliminary potential for the preoperative prediction of malignant risk in thyroid follicular tumors. However, because of the lack of external validation, these findings should be considered hypothesis-generating. The model requires validation in independent cohorts before clinical application.