Clinical AI in Radiology: Foundations, Trends, Applications, and Emerging Directions

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Artificial intelligence (AI) is revolutionizing radiology by enhancing diagnoses, optimizing workflows, and boosting operational efficiency.
• Despite AI's potential, challenges such as data privacy, clinical usability, and workflow integration persist.

Purpose of the Study:

• This review provides context on clinical AI in radiology, discussing foundations and current trends.
• It highlights translational applications and emerging opportunities for AI in oncology radiology.

Main Methods:

• Review of current AI applications in radiology, including local large language model (LLM) deployment for report streamlining.
• Description of multimodal AI frameworks for early disease detection (e.g., cachexia in pancreatic cancer).
• Discussion of privacy-preserving federated learning (FL) and uncertainty-aware de-identification pipelines for secure data handling.

Main Results:

• Local LLM deployment improves radiology report clarity and consistency.
• Multimodal AI shows potential for early detection of conditions like pancreatic cancer-associated cachexia.
• Federated learning and secure de-identification pipelines facilitate collaborative AI development and data sharing.

Conclusions:

• Successful AI translation in oncology radiology requires local deployment, multimodal reasoning, privacy preservation, and human-in-the-loop oversight.
• Emerging AI opportunities include tumor board decision support, clinical trial matching, and quality assurance.