Large language models as an academic resource for radiologists stepping into artificial intelligence research

Area of Science:

• Medical Imaging and Radiology
• Artificial Intelligence in Healthcare
• Machine Learning and Deep Learning

Background:

• Radiologists increasingly use AI for diagnostics and workflow optimization.
• Limited technical expertise hinders AI/ML/DL algorithm application by many radiologists.
• Large language models (LLMs) like GPT-4o can act as virtual advisors for AI in research.

Purpose of the Study:

• Evaluate GPT-4o's effectiveness as a recommender system for AI algorithms in radiology research.
• Assess GPT-4o's ability to enhance radiologists' understanding and implementation of AI tools.
• Determine if LLMs can bridge the knowledge gap in AI for radiology researchers.

Main Methods:

• GPT-4o recommended ML and DL algorithms based on researcher-provided details (dataset characteristics, modality, size, objectives).
• The model served as a virtual advisor, guiding algorithm selection for specific research needs.
• Recommendations were systematically evaluated for clarity, task alignment, model diversity, and baseline selection.

Main Results:

• GPT-4o effectively recommended appropriate ML/DL algorithms for radiology tasks (segmentation, classification, regression).
• Diverse algorithms (e.g., U-Net, Random Forest, EfficientNet) were suggested, aligning with accepted practices.
• The model provided clear and relevant recommendations for AI implementation in medical imaging research.

Conclusions:

• GPT-4o shows promise as a tool for providing AI/ML algorithm recommendations to radiologists and researchers.
• LLMs can democratize AI access in radiology, fostering innovation and improving research quality.
• Further research should explore LLM integration into clinical workflows and professional development.