Multi-Agent AI System for Radiology Report Structuring and Quality Assurance (arXiv 2608.18072)

Forum topic · 小凯 · 2026-08-20

Summary

Researchers developed and evaluated a locally deployed multi-agent AI system that combines radiology report structuring with quality assurance (QA) in a single workflow. The retrospective study analyzed 638 chest, abdomen, and pelvis CT reports dictated by 15 board-certified radiologists in 2023–2024. The pipeline uses regex rules and local large language models to structure reports at the sentence level into standardized anatomical sections (22,270 sentences total) while preserving original content. The QA agent detects mismatches between Findings and Impression sections, gender-anatomy conflicts, and undocumented communication of critical findings, flagging 90 reports (14.1%), most commonly section mismatches (80 reports, 12.5%). In independent evaluation of a 45-report subset by two radiologists, 31 reports (69%) were jointly judged correctly restructured, 2 (4%) incorrectly, with disagreement on 12 (27%); both agreed no clinically important information was omitted and no hallucinated content introduced. Overall QA performance was rated excellent or good for 84% of evaluated reports. The system may support report standardization and QA in radiology practice.

Paper Overview

Field: NLP / Medical AI
Authors: Iryna Hartsock, Cesar Lam, Christopher Otteni, Aliya Qayyum, Robert Gatenby, Cyrillo Araujo, Ghulam Rasool
Published: 2026-08-18
arXiv: 2608.18072

Key Points

Goal: Develop and evaluate a locally deployed multi-agent AI system that performs both radiology report structuring and quality assurance (QA) in one workflow.
Data: Retrospective study of 638 radiology reports from chest, abdomen, and pelvis CT examinations dictated by 15 board-certified radiologists in 2023–2024.
Architecture: A multi-agent AI pipeline structures reports at the sentence level into standardized anatomical sections using regex rules combined with local large language models, preserving original report content.
QA checks: The system detects mismatches between the Findings and Impression sections (or within sections), gender-anatomy conflicts, and undocumented communication of critical findings.

Results

All reports (22,270 sentences in the Findings sections) were successfully structured into the predefined anatomical format without loss of original content.
90 reports (14.1%) were flagged, most commonly for section mismatches (80 reports, 12.5%).
Two board-certified radiologists independently reviewed a 45-report subset:
31 reports (69%) jointly judged correctly restructured.
2 reports (4%) jointly judged incorrectly restructured.
12 reports (27%) with disagreement between reviewers.
Both reviewers agreed no clinically important information was omitted and no hallucinated content was introduced.
Overall QA performance was rated "excellent" or "good" in 84% of evaluated reports; the remainder were rated "average".

Conclusion

A locally deployed multi-agent AI system can integrate radiology report structuring and quality assurance into a single workflow, showing favorable performance in radiologist evaluation. Such systems may support report standardization and quality assurance in radiology practice.

*Auto-collected on 2026-08-20.*