Related Experiment Video
Updated: Jul 16, 2026
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Diagnostic Performance of the ACR TI-RADS, C-TIRADS, K-TIRADS, and EU TI-RADS Classification Systems for Thyroid
Jun-Chao Zhang1, Hong-Lan Feng2, Qi-Yuan Li3
1Department of Ultrasonography, Affiliated Hospital of Jiangnan University, Wuxi, Jiangsu, People's Republic of China.
Objective:
To evaluate and contrast how effectively the ACR TI-RADS, C-TIRADS, K-TIRADS, and EU TI-RAD identify thyroid micronodules.
Methods:
In this prospective study, thyroid micronodules (maximum diameter ≤10 mm) confirmed by postoperative histopathology were enrolled. These nodules were assessed on ultrasound and categorized using ACR TI-RADS, C-TIRADS, K-TIRADS, and EU TI-RADS. Postoperative pathological results served as reference standard. Receiver operating characteristics (ROC) were performed to define optimal cutoff for each TI-RADS classifications in identifying papillary thyroid microcarcinoma, and the subsequent diagnostic efficiencies of the systems were compared.
Results:
The areas under the curve (AUCs) corresponding to ACR TI-RADS, C-TIRADS, K-TIRADS, and EU TI-RADS were 0.685, 0.691, 0.673, and 0.608, respectively. ACR TI-RADS, C-TIRADS, and K-TIRADS did not differ significantly from one another, and all three exceeded the performance of EU TI-RADS. When the Youden index reached maximum, the optimal cut-off points were >Category 4, >Category 4B, >Category 4, and >Category 4, respectively. Under these optimal thresholds, the ACR TI-RADS demonstrated sensitivity, specificity, positive predictive value, negative predictive value, and accuracy values of 90.12%, 46.62%, 90.95%, 44.23%, and 83.86%; those of the C-TIRADS were 85.58%, 48.31%, 90.78%, 36.02%, and 80.21%; those of the K-TIRADS were 88.19%, 45.95%, 90.66%, 39.53%, and 82.11%; and those of the EU TI-RADS were 98.69%, 22.97%, 88.40%, 74.73%, and 87.80%. Overall, all four TI-RADS exhibited relatively high sensitivity, positive predictive value, and accuracy. However, the specificity and negative predictive value remained low.
Conclusion:
The ACR TI-RADS, C-TIRADS, K-TIRADS, and EU TI-RADS demonstrated relatively limited diagnostic performance for thyroid micronodules. However, all thyroid micronodules from a surgically selected cohort with a very high malignancy rate, which is not representative of real-world thyroid nodule populations. The diagnostic metrics and optimal cutoff values cannot be directly extrapolated to the general population of patients with incidentally detected subcentimeter thyroid nodules.
