Comparison of British Thyroid Association, American College of Radiology TIRADS and Artificial Intelligence TIRADS with histological correlation: diagnostic performance for predicting thyroid malignancy and unnecessary fine needle aspiration rate

Abstract

Objectives: To compare diagnostic performance of British Thyroid Association (BTA), American College of Radiology Thyroid Imaging Reporting and Data System (ACR-TIRADS) and Artificial Intelligence TIRADS (AI-TIRADS) for thyroid nodule malignancy. To determine comparative unnecessary fine needle aspiration (FNA) rates. Methods: 218 thyroid nodules with definitive histology obtained during 2017 were included. Ultrasound images were reviewed retrospectively in consensus by two subspecialist radiologists, blinded to histopathology, and nodules assigned a BTA, ACR-TIRADS and AI-TIRADS grade. Nodule laterality and size were recorded to allow accurate histopathological correlation and determine which nodules met criteria for FNA. Results: 77 (35.3%) nodules were malignant. Deeming ultrasound Grade 4–5 as test-positive and 1–2 as test-negative, sensitivity and specificity for BTA was 98.28 and 42.55%, for ACR-TIRADS: 95.24 and 40.57% and for AI-TIRADS: 93.44 and 45.71%. FNA was indicated in 101 (71.6%), 67 (47.5%) and 65 (46.1%) benign nodules utilising BTA, ACR-TIRADS and AI-TIRADS respectively. The unnecessary FNA rate was significantly higher with BTA (46.3%) compared to ACR-TIRADS (30.7%) and AI-TIRADS (29.8%) p < 0.001. Conclusion: BTA, ACR-TIRADS and AI-TIRADS had similar diagnostic performance for predicting thyroid nodule malignancy with sensitivity >93% for all systems when considering ultrasound Grade 4–5 as malignant and Grade 1–2 as benign. ACR-TIRADS and AI-TIRADS both had a significantly lower rate of recommended FNA in benign nodules compared to BTA. Advances in knowledge: BTA, ACR-TIRADS and AI-TIRADS have comparable diagnostic performance with high sensitivity but relatively low specificity for predicting thyroid nodule malignancy in this cohort using histology as gold-standard. Using Grade 1–2 as benign and 4–5 as malignant there were more false negatives with TIRADS but this improved when taking other features into account while BTA had a significantly higher rate of unnecessary FNA.