Large language models for disease diagnosis: a scoping review.

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Informatics
• Computational Pathology

Background:

• Automatic disease diagnosis is crucial for clinical practice.
• Large language models (LLMs) represent a significant advancement in artificial intelligence, demonstrating potential in diagnostic applications.
• A comprehensive understanding of LLM utilization in disease diagnosis is currently lacking.

Purpose of the Study:

• To conduct a comprehensive review of existing literature on LLM-based disease diagnosis methods.
• To analyze the application of LLMs across different disease types, clinical specialties, and data modalities.
• To identify and evaluate the LLM techniques and assessment strategies employed in diagnostic tasks.

Main Methods:

• Systematic literature review of studies employing LLMs for disease diagnosis.
• Categorization of reviewed studies based on disease characteristics, clinical data types, LLM architectures, and evaluation metrics.
• Analysis of trends, gaps, and limitations in the current research landscape.

Main Results:

• Identification of diverse applications of LLMs in diagnosing various diseases across multiple clinical specialties.
• Characterization of the types of clinical data utilized with LLMs for diagnostic purposes.
• Summary of prevalent LLM techniques and evaluation methodologies in the field.

Conclusions:

• LLMs show significant potential for enhancing automatic disease diagnosis.
• Recommendations are provided for the effective application and rigorous evaluation of LLMs in clinical diagnostic settings.
• Future research directions are outlined to address current limitations and advance the field.