Related Experiment Video
Updated: Nov 7, 2025
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Comparison of 5 Thyroid Ultrasound Stratification Systems for Differentiation of Benign and Malignant Nodules and to
Bekir Kuru1, Mehmet Kefeli2, Murat Danaci3
1Department of General Surgery, Ondokuz Mayis University School of Medicine, Samsun, Turkey.
Objective:
We aimed to compare the thyroid ultrasound risk stratification systems (RSSs) of the American College of Radiology (ACR) Thyroid Imaging Reporting and Data System (TI-RADS), European TI-RADS, Korean TI-RADS, and American Thyroid Association (ATA), American Association of Clinical Endocrinologists, American College of Endocrinology, and Associazione Medici Endocrinologi guidelines to differentiate benign from malignant thyroid nodules and to avoid unnecessary fine needle aspiration (FNA).
Methods:
The records of 1143 nodules ≥1 cm that underwent FNA biopsy and thyroidectomy between 2012 and 2020 at our institution were reviewed. Ultrasound categories and FNA recommendation indications of 5 international RSSs were compared with histopathological findings as benign or malignant. The ultrasound categories and recommended FNA indications, the proportion of the avoidable FNA procedures, and false negative rates (FNRs) by different systems were compared with each other.
Results:
Of the 1143 nodules, 45% had thyroid malignancy. FNA recommendation and ultrasound risk classification of ATA guidelines had the highest area under curves of 0.619, and 0.715, respectively. ACR TI-RADS, American Association of Clinical Endocrinologists/American College of Endocrinology/Associazione Medici Endocrinologi guidelines, European TI-RADS, ATA guidelines, and Korean TI-RADS would have avoided FNA for 34.7%, 31%, 25.7%, 20%, and 6% of nodules with an FNR of 24%, 28.5%, 22%, 7.2%, and 1.9%, respectively.
Conclusion:
Our findings showed that all RSSs classified the nodules appropriately for malignancy. ATA guidelines had the highest area under curves and a low FNR, whereas ACR TI-RADS would have spared more patients from FNA with a high FNR.
