## Multi‑Agent AI for Radiology Report Structuring

Multi‑Agent AI for Radiology Report Structuring

A new AI system was built to structure radiology reports and check their quality.

The system runs locally and uses several agents that talk to each other.

One agent reads the raw report.

A second agent extracts key findings and measurements.

A third agent compares the extracted data with a set of quality rules.

If a rule is broken, the system flags the report for review.

The study tested the system on 638 CT reports of the chest, abdomen, and pelvis.

Reports came from 15 board‑certified radiologists.

Results showed higher consistency and fewer missing items than manual checks.

# Simple outline of the multi‑agent flow
def process_report(text):
    raw = ReaderAgent(text)          # read dictation
    data = ExtractorAgent(raw)       # pull findings
    flags = QAAgent(data)            # run quality rules
    return data, flags

The evaluation was done by independent radiologists who did not build the system.

They rated the AI‑generated structures and the flagged items.

Overall scores were better than the original reports.

The work suggests that locally deployed, multi‑agent AI can improve radiology reporting without sharing patient data.

Source: Multi‑Agent AI System for Radiology Report Structuring and Quality Assurance with Independent Radiologist Evaluation