Research map / Multi-agent systems and games / LLM agents

LungNoduleAgent: A Collaborative Multi-Agent System for Precision Diagnosis of Lung Nodules

AAAI 2026 · Multi-agent systems and games › LLM agents · approach: Agent data synthesis and behavior studies · 1 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes LungNoduleAgent, a multi-agent pipeline (Nodule Spotter with MoE detectors and VLM judging panel, Simulated Radiologist with focal prompting, and a Doctor Agent System with graph RAG) for lung nodule CT report generation and malignancy grading.

Problem

Accurate lung nodule morphology description and malignancy grading from CT scans by general and medical VLMs are unreliable, lacking fine-grained perception and pathology knowledge.

Approach

Decomposes nodule diagnosis into detection (MoE detectors with clustering and VLM judging), region-focused report generation (focal prompting with MedPrompt), and multi-agent discussion grounded in a graph-based pathology knowledge base with shared memory.

Key results

• On PrivateA malignancy grading, LungNoduleAgent reaches 86.7% accuracy and 0.889 F1, versus 62.3% for MedGemma-27B (best baseline).
• On LIDC-IDRI, LungDLC-score rises to 83.5 for LungNoduleAgent, against 75.2 for MedGemma-27B.
• Ablation: adding Mask Clustering and Judge Panel raises detection mAP from 67.1% to 79.3%.

Surprising findings

• GPT-4o achieves higher LLM-judge scores than LungNoduleAgent on some datasets, which the authors attribute to evaluator bias.

Limitations

• Two of the three datasets are private and not released, limiting reproducibility.
• Reliance on GPT-4o as LLM judge for report scoring may introduce evaluator bias.

Experimental setup

Datasets: LIDC-IDRI PrivateA PrivateB

Benchmarks: LIDC-IDRI PrivateA PrivateB

Metrics: LLM-score LungDLC-score Pos QA Neg QA Accuracy F1 mAP

Baselines: GPT-4o Claude 3.7 Sonnet Qwen2.5-VL-7B InternVL3-8B LLaVA3.2-11B MedGemma-27B MedR1-3B MedAgent-Pro MedAgents MDAgent LLaVA-Med

Builds on: GraphRAG Describe Anything Model DBSCAN

Closest papers by meaning

• TumorChain: Interleaved Multimodal Chain-of-Thought Reasoning for Traceable Clinical Tumor Analysis (ICLR 2026 · similarity 0.87)
• 3DMedAgent: Unified Perception-to-Understanding for 3D Medical Analysis (ICML 2026 · similarity 0.85)
• CXReasonBench: A Benchmark for Evaluating Structured Diagnostic Reasoning in Chest X-rays (NeurIPS 2025 · similarity 0.84)
• CPathAgent: An Agent-based Foundation Model for Interpretable High-Resolution Pathology Image Analysis Mimicking Pathologists' Diagnostic Logic (NeurIPS 2025 · similarity 0.84)
• Cross-Modal Interactive Perception Network with Mamba for Lung Tumor Segmentation in PET-CT Images (CVPR 2025 · similarity 0.83)
• GEMA-Score: Granular Explainable Multi-Agent Scoring Framework for Radiology Report Evaluation (AAAI 2026 · similarity 0.83)