English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Can Agentic Systems Close the Gap in Radiology with Production AI Tools?

Forum topic · 小凯 · 2026-11-09

Summary

This arXiv paper (2611.09836) by Aubin Charley, Julien Khlaut, Charles Corbière, Pierre Manceron, and Corentin Dancette examines whether agentic systems can overcome the limitations of current radiology AI. Radiology AI has evolved along two paths: specialized models that excel on narrow tasks but cover few findings, and generalist models with broader coverage but prone to hallucination and inaccurate measurements. The authors propose a radiology agentic framework in which an orchestrator calls modular tools to produce structured reports, evaluated using tools derived from FDA-cleared radiology AI devices. Report generation is simulated on a large multi-modality report corpus, comparing a catalog restricted to cleared-device finding classes against an unrestricted catalog, with planning driven by clinical indications or target reports. With the cleared catalog and indications only, the orchestrator recovered under one-third of clinically actionable findings; recovery improved substantially with the unrestricted catalog and further with target-report planning. Most missed findings stemmed from classes not offered by cleared devices for the relevant modality, and one-third from available tools that were never invoked. Adding the most frequently missing tools closed most of the gap on held-out reports, while a smaller orchestrator with the unrestricted catalog forfeited all gains. Tool availability and effective orchestration emerge as complementary constraints.

Paper Overview

  • Research area: Machine Learning
  • Authors: Aubin Charley, Julien Khlaut, Charles Corbière, Pierre Manceron, Corentin Dancette
  • Published: 2026-11-06
  • arXiv: 2611.09836
  • Summary

    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 toward comprehensive radiology reporting, but published systems are tied to one modality or body part.

    This paper describes a radiology agentic framework that standardizes how an orchestrator calls modular tools to produce a structured report, and uses it to measure what such a system can do with tools from FDA-cleared radiology AI devices.

    Methodology

  • Report generation is simulated on a large multi-modality report corpus.
  • The orchestrator is given either:
  • (a) a catalog of finding classes consolidated from cleared devices (each available on its cleared modality), or
  • (b) an unrestricted catalog where any class can be requested.
  • Planning is driven by clinical indications or by a target report.
  • Tool outputs are simulated from reference reports, so the comparison isolates tool availability and orchestration, excluding tool error.
  • Key Findings

  • With the cleared catalog and indications only, the orchestrator recovered less than one-third of clinically actionable findings.
  • Recovery improved significantly with the unrestricted catalog, and further when planning came from a target report.
  • Attributing each missed finding to a cause showed that most were classes not offered by cleared devices on the study's modality; one-third were classes or tools that were available but never invoked.
  • Adding the most frequently missing tools closed most of the gap with the unrestricted catalog on held-out reports.
  • A smaller orchestrator using the unrestricted catalog forfeited all of the gains.

Conclusion

Tool availability and effective orchestration are two complementary constraints on comprehensive AI-assisted radiology reporting.

---

*Auto-collected on 2026-11-10*

Tags

#radiology-ai#agentic-systems#machine-learning#medical-imaging#arxiv#report-generation#fda-cleared-devices

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178636993