Building radiomics models based on ACR TI-RADS combining clinical features for discriminating benign and malignant thyroid nodules
PurposeThe aim of this study was to establish and validate a radiomics model combining the American College of Radiology Thyroid Imaging, Reporting and Data System (ACR TI-RADS) and clinical features and to build a nomogram that could be utilized to enhance the diagnostic performance of malignant th...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2025-07-01
|
| Series: | Frontiers in Endocrinology |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fendo.2025.1486920/full |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | PurposeThe aim of this study was to establish and validate a radiomics model combining the American College of Radiology Thyroid Imaging, Reporting and Data System (ACR TI-RADS) and clinical features and to build a nomogram that could be utilized to enhance the diagnostic performance of malignant thyroid nodules.MethodFrom January 2019 to September 2022, 329 thyroid nodules from 323 patients who had been referred for surgery and had pathological evidence of them were gathered retrospectively and randomly allocated to training and test cohorts (8:2 ratio). A total of 107 radiomics features were extracted from the US images, and the radiomics score (Rad-score) was constructed using the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm. Different models were created using logistic regression, including the clinic-ACR score (Clin+ACR), clinic-Rad score (Clin+Rad), ACR score-Rad score (ACR+Rad), and combined clinic-ACR score-Rad score (Clin+ACR+Rad). The diagnostic performance of different models was calculated and compared using the area under the receiver operating curve (AUC) and the corresponding sensitivity and specificity.ResultsEight radiomics features were independent signatures for predicting malignant TNs, with malignant TNs having higher Rad-scores in both cohorts (P < 0.05). The Clin+ACR+Rad model showed excellent diagnostic prediction ability in both the training (AUC = 0.958) and test datasets (AUC = 0.937), significantly outperforming other models including Rad-score (AUC = 0.890, 0.856), Clin+Rad (AUC = 0.895, 0.859), ACR+Rad (AUC = 0.943, 0.934), and Clin+ACR (AUC = 0.784, 0.785) (all P < 0.05). The calibration curve demonstrated that the mean absolute error in the training group was just 0.020 and in the test cohort was 0.033. To evaluate the clinical utility of the nomogram in reducing unnecessary biopsies, we further analyzed the performance of our integrated model (Clin+ACR+Rad) compared to the traditional ACR TI-RADS system at different probability thresholds. At the statistically optimal threshold of 0.386, the unnecessary biopsy rate decreased from 46.97% to 22.05% in the training cohort and from 45.83% to 21.05% in the test cohort.ConclusionThe current study offers preliminary support that the model of combined clinic-ACR score-radiomics score can be helpful for predicting malignancy in thyroid nodules by looking at a retrospective cohort of surgically treated thyroid nodules. The Clin-ACR-Rad nomogram may be a more practical instrument and more accurate prediction model for malignant thyroid nodules. |
|---|---|
| ISSN: | 1664-2392 |