Foundation models for radiology: fundamentals, applications, opportunities, challenges, risks, and prospects

Area of Science:

• Artificial Intelligence (AI)
• Medical Imaging
• Radiology

Background:

• Foundation models (FMs) represent a significant advancement over traditional narrow AI.
• FMs offer enhanced adaptability, multimodal integration, and generalizability in diverse fields, including radiology.

Purpose of the Study:

• To provide a comprehensive overview of FMs in radiology.
• To discuss the applications, opportunities, and challenges of FMs in clinical practice.
• To guide the responsible adoption of FMs in healthcare.

Main Methods:

• Literature review and synthesis of current research on FMs in radiology.
• Analysis of the potential applications of FMs, such as image interpretation and report generation.
• Identification and discussion of challenges related to clinical translation, including ethical and regulatory concerns.

Main Results:

• FMs demonstrate potential for diverse applications like integrative diagnostics and synthetic data creation.
• Key challenges include model opacity, data privacy, ethical considerations (bias, fairness), and regulatory uncertainty.
• Rigorous validation is crucial to address stochasticity and potential hallucination.

Conclusions:

• FMs hold significant promise for advancing precision medicine and healthcare.
• Addressing challenges is essential for the responsible and effective clinical translation of FMs in radiology.
• This overview aims to facilitate informed decision-making regarding FM adoption.