Value of Intratumoral and Peritumoral Ultrasound Radiomics Nomogram for Differential Diagnosis of C‐TIRADS 4 Thyroid Nodules
Kairen Zhang, Dan Zhao, XiaoFeng Wu, Chenyang Jin, Ying Song, Fenglin DongABSTRACT
Objective
The Chinese Thyroid Imaging Reporting and Data System (C‐TIRADS) 4 thyroid nodules (TNs) have intricate sonographic features that potentially lead to overdiagnosis or misdiagnosis. This study aimed to improve the precision of differentiating C‐TIRADS 4 TNs using radiomic nomograms that include both intratumoral and peritumoral features.
Methods
A total of 337 patients with C‐TIRADS 4 thyroid nodules confirmed by pathology or FNAB were retrospectively enrolled and randomly divided into a training cohort ( n = 269) and a validation cohort ( n = 68) at a ratio of 8:2. Radiomics features were extracted from intratumoral and peritumoral (1–3 mm) regions on B‐mode ultrasound images. Feature stability was assessed using intraclass and intraclass correlation coefficients (ICC). Key features were selected through a sequential process involving the Mann–Whitney U test, recursive feature elimination (RFE), and least absolute shrinkage and selection operator (LASSO) regression, followed by calculation of the radiomics score (Rad‐score). Independent predictors were identified by univariate and multivariate logistic regression. Three models were constructed: a clinical and ultrasound model (Model 1), a radiomics model (Model 2), and a combined nomogram integrating both (Model 3). Model performance and clinical utility were evaluated using receiver operating characteristic curves, calibration curves, and decision curve analysis (DCA).
Results
Multivariate analysis showed that age, margin, echogenic foci, and the intratumoral and peritumoral 1 mm Rad‐score were independent predictors for the benign or malignant nature of C‐TIRADS 4 nodules ( p < 0.05). The combined nomogram (Model 3) achieved an area under the curve (AUC) of 0.852 (95% CI: 0.804–0.892) in the training cohort, significantly higher than that of Model 1 (AUC = 0.737) and Model 2 (AUC = 0.784) ( p < 0.05). In the validation cohort, Model 3 yielded an AUC of 0.916 (95% CI: 0.851–0.980), outperforming Model 1 (AUC = 0.766) and Model 2 (AUC = 0.807) (both p < 0.05). The calibration curve of the nomogram showed good agreement with the ideal curve (Hosmer‐Lemeshow test p > 0.05), and DCA indicated that it provided the greatest clinical net benefit.
Conclusion
The nomogram integrating clinical characteristics, ultrasound characteristics, and intratumoral and peritumoral 1 mm Rad‐score can distinguish benign from malignant C‐TIRADS 4 TNs, with superior diagnostic performance compared to single‐modality models. It may serve as a non‐invasive and reliable tool to reduce unnecessary FNAB rates.