Comparison of the diagnostic performance of the artificial intelligence-based TIRADS algorithm with established...

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Thyroid nodules are common, and accurate risk stratification is crucial for appropriate management.
• Various Thyroid Imaging Reporting and Data Systems (TIRADS) exist to standardize nodule assessment.
• The diagnostic performance of newer systems, like AI-TIRADS, requires thorough evaluation.

Purpose of the Study:

• To compare the diagnostic performance of multiple TIRADS classifications, including AI-TIRADS, for thyroid nodules.
• To assess the effectiveness of AI-TIRADS in improving specificity and reducing unnecessary biopsies.
• To evaluate the impact of suspicious ultrasound features on malignancy risk.

Main Methods:

• Retrospective analysis of 1,322 thyroid nodules from 1,139 patients with cytopathological diagnoses.
• Assessment of nodules using ACR-TIRADS, ATA-TIRADS, EU-TIRADS, K-TIRADS, and AI-TIRADS by three independent radiologists.
• Evaluation of diagnostic performance metrics including sensitivity, specificity, PPV, and NPV, with McNemar test for comparisons.

Main Results:

• AI-TIRADS demonstrated superior specificity (53.6%) compared to ACR-TIRADS (44.6%), ATA-TIRADS (39.3%), EU-TIRADS (40.1%), and K-TIRADS (40.1%) (P < 0.001).
• AI-TIRADS maintained high sensitivity without compromising diagnostic accuracy.
• The presence of multiple suspicious ultrasound features significantly increased malignancy risk.

Conclusions:

• AI-TIRADS shows significant promise for enhancing thyroid nodule risk stratification by improving diagnostic specificity.
• AI-TIRADS may lead to a reduction in unnecessary thyroid biopsies and optimize patient management.
• The study highlights the potential of AI-TIRADS in improving the efficiency of healthcare resource utilization in thyroid nodule assessment.