ObjectiveTo develop a nomogram model combining ultrasound radiomics features and clinical features, and to evaluate its predictive value for pathological invasiveness of papillary thyroid carcinoma (PTC).MethodsThis study included 224 patients diagnosed with PTC between January 2024 and May 2025 from Zhejiang Provincial People's Hospital. Relevant laboratory data and raw ultrasound images were collected. The patients were randomly divided into training and testing cohorts at an 8:2 ratio. The study additionally collected 42 patients diagnosed with PTC from June to November 2025 as an independent validation cohort. Pathologically invasive positivity was defined as the presence of one or more of the following features: extrathyroidal extension, vascular invasion, perineural invasion, intraglandular dissemination, extra glandular invasion, central or lateral cervical lymph node metastasis, or high-risk subtypes. A univariate analysis was conducted on 11 candidate clinical features. Then, stepwise logistic regression was employed to identify independent predictors of pathological invasiveness in PTC and to develop a predictive model. To screen for radiomic features, we applied the Mann-Whitney U test and Spearman's correlation analysis with a threshold of 0.9, in conjunction with greedy recursive pruning and least absolute shrinkage and selection operator (LASSO) regression. The selected features were subsequently input into eight machine learning algorithms: logistic regression, support vector machines, k-nearest neighbors, extreme random trees, random forests, XGBoost, LightGBM, and multi-layer perceptrons, to construct predictive models. The optimal algorithm was determined based on the area under the curve (AUC) metric. By integrating the clinical feature model and the radiomics model into a logistic regression framework, a receiver operating characteristic (ROC) curve was generated. The overall discriminatory power of the nomogram model, its calibration accuracy, and the clinical utility across various probability thresholds were assessed using ROC curves, calibration plots, and decision curve analysis (DCA) curves.ResultsA univariate analysis indicated that thyroid nodules in the pathologically invasive group were significantly larger compared with those in the pathologically non-invasive group (all P ConclusionsA nomogram model combining ultrasound radiomics and clinical features was successfully developed and validated, enabling non-invasive and quantitative prediction of pathological invasiveness in PTC. The model demonstrated good discrimination, calibration, and clinical utility.
LIU et al. (2026) studied this question.