Performance of Artificial Intelligence Versus Radiologists in Acr Ti-Rads Stratification of Thyroid Nodules: An Ultrasound-Based Comparative Study

Objective To evaluate the diagnostic performance of a computer-aided diagnosis (CAD) system compared with radiologist assessment for ACR TI-RADS classification of thyroid nodules on ultrasound. Methods This retrospective cross-sectional study included 332 thyroid nodules from 306 patients. Seven radiologists with different levels of experience (junior and senior) independently classified nodules according to the five ACR TI-RADS ultrasound features. The same images were analyzed using the CAD software AmCAD-UT™ Detection. Diagnostic performance was assessed using sensitivity, specificity, positive predictive value, negative predictive value, accuracy, positive likelihood ratio, and negative likelihood ratio. Results Senior radiologists achieved the highest specificity and overall accuracy, with sensitivity of 57.4%, specificity of 80.0%, and accuracy of 69.7%, whereas junior radiologists showed sensitivity of 66.7%, specificity of 64.2%, and accuracy of 64.5%. The AI system showed sensitivity of 53.6% and specificity of 68.4%. Conclusion The evaluated AI system does not outperform radiologist assessment in ACR TI-RADS classification of thyroid nodules and cannot replace human expertise. However, it may serve as a complementary tool, particularly as a second opinion or training support for less experienced radiologists.