Systematic Review: Agentic AI in Neuroradiology: Technical Promise with Limited Clinical Evidence.

Area of Science:

• Artificial Intelligence
• Neuroradiology
• Medical Imaging

Background:

• Agentic artificial intelligence (AI) systems, incorporating iterative reasoning and autonomous tool use, are proposed to overcome limitations of large language models (LLMs) in neuroradiology.
• The clinical implementation and validation of these agentic AI systems in neuroradiology remain largely unassessed.

Purpose of the Study:

• To systematically review and assess the implementation and clinical validation of agentic AI in neuroradiology.
• To evaluate the evidence scarcity, methodological rigor, and clinical utility of agentic AI applications in the field.

Main Methods:

• A systematic literature search was conducted across PubMed, Web of Science, and Scopus (January 2022-August 2025).
• Studies were included if they implemented agentic AI, defined as requiring iterative reasoning plus autonomous tool use or multi-agent collaboration.
• Six independent reviewers assessed study quality using adapted QUADAS-AI criteria, focusing on implementation, validation, and outcomes.

Main Results:

• Only 9 out of 230 records (3.90%) met inclusion criteria, indicating severe evidence scarcity.
• A significant portion of studies (30%) misrepresented their AI as agentic, lacking genuine autonomy or multi-agent collaboration.
• The sole randomized controlled trial demonstrated high technical performance but no measurable clinical benefit, highlighting a gap between technical accuracy and clinical utility. Safety assessments were universally absent.

Conclusions:

• Agentic AI in neuroradiology is technically promising but clinically unproven, remaining in its early research phase.
• Current evidence is insufficient to support clinical deployment due to methodological limitations and a lack of demonstrated clinical utility.
• Rigorous, multi-center prospective trials focusing on patient-centered and safety outcomes are essential before responsible clinical implementation can be considered.