Current status of machine learning in thyroid cytopathology

Area of Science:

• Computational Pathology
• Digital Cytopathology
• Artificial Intelligence in Medicine

Background:

• Digital Pathology enables computational approaches, but AI in cytology faces challenges due to technical limitations with specimen scanners.
• FDA- Breakthrough Device Designations have primarily focused on tissue specimens, with less development in cytology.

Purpose of the Study:

• To evaluate the potential of machine learning algorithms (MLA) for decision-support tools in cytopathology, particularly for thyroid fine needle aspiration biopsy (FNAB) specimens.
• To explore the impact of AI-assisted algorithms on the accuracy and efficiency of thyroid cytology diagnosis.

Main Methods:

• Review of studies evaluating machine learning algorithms applied to digital cytology images of thyroid specimens.
• Analysis of the performance of MLA in diagnosing and classifying thyroid cytology samples.

Main Results:

• Promising results indicate increased diagnostic accuracy and classification capabilities for thyroid cytology using MLA.
• MLA applications offer new insights and demonstrate potential for improving cytopathology workflow efficiency.

Conclusions:

• Machine learning algorithms hold significant potential for improving the speed and accuracy of thyroid cancer diagnosis.
• Larger, multi-institutional datasets are crucial for optimal training and validation of MLA models in thyroid cytology to enhance patient management.