[A nomogram combining ultrasound radiomics and clinical features for predicting pathological invasiveness of...

To develop a nomogram model combining ultrasound radiomics and clinical features and to evaluate its predictive value for pathological invasiveness of papillary thyroid carcinoma (PTC).

This study included 224 patients diagnosed with PTC between January 2024 and May 2025 at Zhejiang Provincial People's Hospital. Baseline clinical data, laboratory findings, and raw ultrasound images were collected. The patients were randomly divided into a training cohort (n=179) and a testing cohort (n=45) at an 8∶2 ratio. Additionally, 42 patients diagnosed with PTC from June to November 2025 were enrolled as an independent validation cohort. Pathological invasiveness 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. Univariate analysis was performed on 11 candidate clinical features, followed by stepwise logistic regression to identify independent predictors and to construct a clinical feature-based model. Ultrasound radiomics features were screened using Mann-Whitney U test, Spearman's correlation analysis (threshold 0.9) with greedy recursive elimination, and least absolute shrinkage and selection operator (LASSO) regression. The selected features were then input into eight machine learning algorithms to build predictive models, and the optimal algorithm was selected based on the area under the receiver operating characteristic (ROC) curve (AUC). A nomogram was subsequently constructed by integrating the clinical model and the ultrasound radiomics model into a logistic regression framework. The discriminative ability, calibration, and clinical utility of the nomogram were assessed using ROC curves, calibration curves, and clinical decision curves.

The clinical feature model based solely on nodule size achieved AUCs of 0.889 (95%CI: 0.843-0.935) and 0.934 (95%CI: 0.860-0.973) in the training and testing cohorts, respectively. Thirteen ultrasound radiomics features were selected. The radiomics model built with logistic regression yielded AUCs of 0.846 (95%CI: 0.791-0.902) and 0.939 (95%CI: 0.871-1.000) in the training and testing cohorts, respectively. The nomogram combining ultrasound radiomics and clinical features achieved AUCs of 0.902 (95%CI: 0.858-0.945) in the training cohort, 0.982 (95%CI: 0.953-1.000) in the testing cohort, and 0.803 (95%CI: 0.665-0.941) in the independent validation cohort. Calibration curves demonstrated good agreement between predicted and observed probabilities, and clinical decision curves indicated favorable net clinical benefit.

A nomogram model combining ultrasound radiomics and clinical features was successfully developed, which exhibited good discrimination, calibration, and clinical utility.