ARTIFICIAL INTELLIGENCE IN TIRADS BASED RISK STRATIFICATION OF THYROID NODULES. AN UPGRADE OR ANOTHER BRICK IN THE WALL? shreyamsam@gmail.com

 

ARTIFICIAL INTELLIGENCE IN TIRADS BASED RISK STRATIFICATION OF THYROID NODULES. AN UPGRADE OR ANOTHER BRICK IN THE WALL?

2

 

 

 

 

India

Oral or Poster

Ultrasound based risk-stratification systems for thyroid-nodules(TN) are ever-evolving,with improvements aimed at reducing unnecessary interventions and/or diagnose malignancy for intervention. We aim to compare utility,diagnostic accuracy of American College of Radiology-Thyroid Imaging Reporting and Data System(ACR-TIRADS) with an artificial-intelligence (AI) optimized version of the same (AI-TIRADS).

Retrospective study,181 TN from 167-patients (January-2023 to March-2025). Only nodules with ultrasonography in accordance with ACR-TIRADS, and final histopathology-report (HPE) included. Fresh scores assigned for nodules and categorized according to AI-TIRADS. Risk-stratification of both compared against HPE and statistical analyses performed. Diagnostic performance assessed by using area under receiver-operating curve.

Mean age- 36.79 +/- 8.97 years. Male: female- 151:16. Mean nodule size was 2.98 +/- 1.09 cm. 139 (76.8%) nodules were benign on final histopathology while 42 (23.2%) were malignant. After assigning risk categories, AI-TIRADS recommended 28 lesser fine-needle aspirations (FNA) than ACR-TIRADS. Only one of these 28 was malignant on HPE (false-negative). The reduction in FNA was clinically significant (p=0.001). AI-TIRADS upstaged risk & recommended FNA for 8 nodules (but ACR-TIRADS did not) & all 8 were malignant (p=0.001). There were no false positives. The sensitivity & specificity of ACR-TIRADS was 71.4% & 74.8% while that of AI-TIRADS was 88.1% & 89.9%. Positive and negative predictive value for ACR-TIRADS was 46.2% and 89.7%, and for AI-TIRADS was 72.5% and 96.2% respectively. The area under curve for ACR-TIRADS was 0.731 and for AI-TIRADS was 0.890.

Optimization with artificial intelligence improves overall performance of ACR-TIRADS by reducing unnecessary FNA & improving detection of malignant nodules.