Agentic systems improve radiology AI by combining tools and planning
Can Agentic Systems Close the Gap in Radiology with Production AI Tools?
Artificial IntelligenceMachine Learning
Summary
AI tools for reading medical images either focus on small tasks very well or try to cover many tasks but often make mistakes. The authors studied a way to combine these tools smartly by using an agent that decides which tool to use for what job in radiology reports. Their system shows that having more available tools and better planning helps cover many important findings in reports. They also found that missing tools and not using available tools properly both limit how well these systems can work.
What this means in practice
- •For radiology ai developers: Build radiology reporting AI that combines specialized tools under a central planner to improve coverage of clinical findings from FDA-cleared modules.$Commercial implications: Enables selling integrated AI reporting platforms that better cover multi-modality radiology findings using proven cleared AI tools.
- •For hospital technology teams: Integrate modular AI tools with a managing agent to enhance AI-assisted radiology reports covering comprehensive clinical findings.
Authors
Aubin Charley, Julien Khlaut, Charles Corbière, Pierre Manceron, Corentin Dancette, Léo Machado
Abstract
Radiology AI has developed along two paths: specialized models that perform well on narrow tasks but cover few clinical findings, and generalist models that address a broader range of tasks but remain prone to hallucination and inaccurate measurements. Combining the two through agentic systems offers a route towards comprehensive radiology reporting, but published systems are tied to one modality or body part. We describe a radiology agentic framework that standardizes how an orchestrator calls modular tools to produce a structured report, and use it to measure what such a system can do with the tools from FDA-cleared radiology AI devices. Report generation is simulated on a large multi-modality report corpus, with the orchestrator given either a catalog of finding classes consolidated from the cleared devices, each available on the modalities it is cleared for, or an unlimited catalog in which any class can be requested, and planning either from the clinical indication or from the target report. Tool outputs are simulated from the reference report, so the comparison isolates tool availability and orchestration from tool error. With the cleared catalog and the indication alone, the orchestrator recovers under a third of the clinically actionable findings; recovery improves substantially with the unlimited catalog, and further when the plan is written from the target report. Attributing every missed finding to a cause shows that most are classes no cleared device offers for the study's modality, and that a third are classes or tools that were offered and not called. Adding the most frequently missing tools closes most of the gap to the unlimited catalog on held-out reports, whereas a smaller orchestrator with the unlimited catalog gives up the whole gain. Tool availability and effective orchestration are thus complementary constraints on comprehensive AI-assisted radiology reporting.