Generative AI and Foundation Models in Radiology: Applications, Opportunities, and Potential Challenges

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Foundation models (FMs) utilize self-attention mechanisms for multimodal data processing.
• FMs can be adapted to specific medical imaging tasks using smaller datasets via transfer learning, fine-tuning, and few-shot learning.
• Generative AI capabilities within FMs aid in creating synthetic medical images, addressing annotation limitations.

Purpose of the Study:

• To review the evolving role of FMs and generative AI in radiology.
• To highlight recent research advances and clinical applications of FMs in medical imaging.
• To identify key challenges for the responsible clinical integration of FMs.

Main Methods:

• Review of current literature on Foundation models in medical imaging.
• Analysis of applications in radiology across various imaging modalities.
• Exploration of adaptation techniques like transfer learning and prompt engineering.
• Examination of generative AI for synthetic data creation.

Main Results:

• FMs show potential to enhance diagnostic accuracy and streamline workflows in radiology.
• Adaptation techniques make FMs valuable in data-scarce medical imaging scenarios.
• Generative AI assists in overcoming data annotation limitations.

Conclusions:

• Clinical integration of FMs in radiology faces challenges including interpretability, bias, privacy, regulations, and computational costs.
• Addressing these barriers requires collaboration between technical developers, healthcare providers, and regulatory bodies.
• Responsible deployment of FMs in radiology necessitates overcoming identified challenges.