Deep-Learning-Based Screening and Ancillary Testing for Thyroid Cytopathology.

Area of Science:

• Endocrinology
• Oncology
• Pathology
• Artificial Intelligence in Medicine

Background:

• Thyroid cancer is the most common endocrine malignancy.
• Fine-needle aspiration biopsies (FNABs) are crucial for preoperative risk assessment.
• Indeterminate FNAB results often lead to diagnostic challenges and unnecessary surgeries.

Purpose of the Study:

• To develop and validate a deep-learning algorithm for analyzing thyroid FNAB whole-slide images (WSIs).
• To assess the algorithm's performance in classifying determinate cases and aiding in the disambiguation of indeterminate cases.

Main Methods:

• Development of a deep-learning algorithm for analyzing thyroid FNAB WSIs.
• Testing on the largest reported dataset of thyroid FNAB WSIs.
• Validation using a separate dataset of consecutive FNABs from a full calendar year.

Main Results:

• The algorithm achieved clinical-grade performance, classifying 45.1% of WSIs as benign or malignant.
• It reduced indeterminate cases by reclassifying 21.3% as benign.
• Results demonstrated clinically acceptable margins of error for thyroid FNAB classification.

Conclusions:

• Deep learning applied to thyroid FNAB WSIs offers a powerful tool for improving diagnostic accuracy.
• The algorithm shows potential as an ancillary test to reduce diagnostic uncertainty and unnecessary surgeries.
• This AI-driven approach can enhance the preoperative risk assessment of thyroid nodules.