Objective:

To summarize the current role and performance of AI tools in the diagnosis of thyroid cancer, particularly in ultrasound evaluation of thyroid nodules, and their significance in improving diagnostic accuracy.

Key Findings:

• AI systems improve diagnostic performance for thyroid nodules, particularly for less experienced physicians.
• FDA-cleared platforms utilize established risk stratification systems and have shown significant improvements in accuracy.
• Large language models exhibit variable performance, particularly in specific contexts, and are not recommended for clinical decision-making.
• AI shows promise in detecting cervical lymph node metastases and in cytology, but no tools are currently approved for these applications.

Interpretation:

AI tools are enhancing the diagnostic process for thyroid cancer, but their integration into clinical practice faces challenges related to workflow, validation, reimbursement, and the need for compelling evidence.

Limitations:

• AI systems have not achieved widespread adoption due to workflow friction, uncertain ROI, and potential biases in studies.
• Lack of independent validation studies in typical clinical settings.
• Current AI tools are primarily designed to augment, not replace, clinical judgment.

Conclusion:

AI tools hold significant promise for improving the evaluation of thyroid nodules and may extend to lymph nodes and biopsy specimens, but successful implementation requires addressing operational and financial barriers.