MARCH: Multi-Agent Radiology Clinical Hierarchy for CT Report Generation

2026-04-20 · Source: cs.AI updates on arXiv.org · Field: Health & Wellbeing — Artificial Intelligence & Machine Learning, Medical Devices & Health Technology, Clinical Care & Medical Practice · Depth: Expert, extended

Summary

MARCH (Multi-Agent Radiology Clinical Hierarchy) is a novel multi-agent framework designed to automate 3D radiology report generation, addressing issues like clinical hallucinations and lack of iterative verification common in existing Vision-Language Models (VLMs). The framework emulates a radiology department's professional hierarchy, assigning specialized roles to distinct agents. It features a Resident Agent for initial drafting using multi-scale CT feature extraction, multiple Fellow Agents for retrieval-augmented revision, and an Attending Agent that orchestrates an iterative, stance-based consensus discourse to resolve diagnostic discrepancies. Evaluated on the RadGenome-ChestCT dataset, MARCH significantly outperforms state-of-the-art baselines in both clinical fidelity and linguistic accuracy, demonstrating that human-like organizational structures can enhance AI reliability in high-stakes medical domains.

Key takeaway

For Computer Vision Engineers developing medical AI, MARCH demonstrates that structuring AI agents to mimic human clinical hierarchies significantly enhances report accuracy and reduces hallucinations. You should consider adopting multi-agent frameworks with iterative consensus and retrieval-augmented revision to improve the reliability and clinical fidelity of your automated diagnostic systems, especially for complex 3D imaging data like CT scans. This approach can lead to more trustworthy and clinically coherent outputs.

Key insights

Modeling human clinical hierarchies with multi-agent AI improves radiology report accuracy and reduces hallucinations.

Principles

• Hierarchical multi-agent systems enhance AI reliability.
• Iterative consensus discourse resolves diagnostic discrepancies.
• Retrieval-augmented generation improves diagnostic grounding.

Method

MARCH employs a three-stage process: initial drafting by a Resident Agent, retrieval-augmented revision by Fellow Agents, and consensus-driven finalization orchestrated by an Attending Agent through iterative discourse.

In practice

• Implement multi-agent systems for complex diagnostic tasks.
• Use retrieval-augmented generation to reduce AI hallucinations.
• Incorporate iterative consensus mechanisms for high-stakes outputs.

Topics

• MARCH Framework
• Radiology Report Generation
• Multi-Agent Systems
• Clinical Hallucinations
• Retrieval-Augmented Generation

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• MARCH: Multi-Agent Radiology Clinical Hierarchy for CT Report Generation — A multi-agent framework mimicking radiology hierarchies improves CT report generation fidelity and accuracy.
• MARCH: Multi-Agent Reinforced Self-Check for LLM Hallucination — MARCH uses multi-agent reinforcement learning and information asymmetry to reduce LLM hallucination in RAG systems.
• Evaluating Multi-Agent LLM Architectures for Rare Disease Diagnosis — Multi-agent LLM architectures do not guarantee improved diagnostic accuracy; topology selection is critical.
• DermAgent: A Self-Reflective Agentic System for Dermatological Image Analysis with Multi-Tool Reasoning and Traceable Decision-Making — DermAgent uses a multi-tool, self-reflective agentic system with external knowledge retrieval to enhance dermatological diagnosis and mitigate MLLM hallucinations.
• MARCUS: An agentic, multimodal vision-language model for cardiac diagnosis and management — MARCUS is an agentic, multimodal vision-language model excelling in cardiac diagnosis by integrating diverse imaging data.
• [R] Toward Guarantees for Clinical Reasoning in Vision Language Models via Formal Verification — Formal verification can mathematically prove VLM diagnostic claims are consistent with stated findings.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.