External validation of AIBx, an artificial intelligence model for risk stratification, in thyroid nodules

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Thyroid nodule diagnosis

Background:

• Artificial intelligence (AI) algorithms offer potential to objectively risk-stratify thyroid nodules, reducing ultrasonography subjectivity.
• The AIBx algorithm has demonstrated promising performance in preliminary studies.
• External validation is essential before widespread clinical adoption of AI tools for thyroid nodule assessment.

Purpose of the Study:

• To externally validate the AIBx artificial intelligence algorithm for thyroid nodule risk stratification.
• To compare the performance of AIBx against histopathology and physician-derived Thyroid Imaging Reporting and Data System (TIRADS) scores.
• To evaluate the combined utility of AIBx and TIRADS in improving diagnostic accuracy for malignant thyroid nodules.

Main Methods:

• Retrospective analysis of ultrasound images from 257 thyroid nodules (1-4 cm) in 209 patients undergoing surgery.
• Exclusion of specific nodule types: medullary thyroid cancer, metastases, lymphomas, and purely cystic lesions.
• Comparison of AIBx algorithm results with histopathology and physician-assessed TIRADS categories.

Main Results:

• AIBx achieved a negative predictive value (NPV) of 89.2% and a sensitivity of 78.4%.
• Combining AIBx with TIRADS (categories 4 and 5) improved NPV to 93.0% and ensured no malignant nodules were missed.
• The combined approach demonstrated superior performance in reducing false negatives compared to TIRADS alone.

Conclusions:

• The AIBx algorithm shows comparable negative predictive values to TIRADS when applied to external datasets.
• Combining AIBx with TIRADS enhances diagnostic accuracy and reduces false-negative assessments for thyroid nodules.
• This AI tool has the potential to assist less experienced clinicians by mitigating the subjectivity of ultrasound interpretation.