Artificial Intelligence in Thyroid Cytopathology: Diagnostic and Technical Insights.

Area of Science:

• Cytopathology
• Artificial Intelligence
• Medical Imaging

Background:

• Fine-needle aspiration cytology (FNAC) is standard for thyroid nodule evaluation using the Bethesda System.
• Indeterminate categories (Bethesda III-IV) present challenges, leading to overtreatment or delayed testing.
• Deep learning (DL) offers potential applications in thyroid cytopathology, including triage and classification.

Purpose of the Study:

• To provide a technical overview of DL in thyroid cytology.
• To highlight challenges and prerequisites for clinical deployment.
• To discuss future clinical applications of DL in this field.

Main Methods:

• Review of DL applications in thyroid cytopathology.
• Emphasis on preanalytical variability, architectural choices, and clinical integration.
• Discussion of requirements for safe clinical implementation.

Main Results:

• DL shows promise for triage, Bethesda classification, and molecular data integration.
• Preanalytical variability, annotation bias, and domain shift limit generalizability.
• Most current studies are retrospective and lack external validation.

Conclusions:

• Standardized datasets, multicenter trials, and explainability are crucial for DL deployment.
• DL is likely to serve as diagnostic co-pilots and risk stratification tools.
• Validated DL can augment cytopathologists, reduce variability, and drive a data-driven discipline.