AI Chest X-Ray Agent
Multimodal AI agent unifying 8 specialized tools and LLMs into one framework. Detect 18 pathologies, segment organs, generate reports — published at ICML 2025.
Upload Your Chest X-Ray
Get AI-powered pathology detection, organ segmentation, and radiology reports in seconds
Drag your X-ray image here
or click to select files
Supports JPEG, PNG, WebP • Max 10MB
Classification
Detect 18 pathologies
Segmentation
Analyze organ regions
Report
Generate medical reports
How Does This AI Radiology Agent Work?
A multimodal medical AI framework that dynamically selects and sequences specialized chest X-ray analysis tools to answer complex clinical queries.
- Dynamic Tool OrchestrationAn LLM-driven ReAct loop breaks complex medical queries into sequential steps, selecting the right specialized tool for each task.
- 8 Integrated Analysis ToolsCombines DenseNet-121, CheXagent, LLaVA-Med, MedSAM, PSPNet, Maira-2, SwinV2, and RoentGen for comprehensive diagnostics.
- No Retraining RequiredModular architecture allows adding new tools without retraining. Built on LangChain and LangGraph for reliable orchestration.
Why Use This AI Chest X-Ray Analysis Tool?
Proven accuracy on medical benchmarks, with capabilities spanning pathology detection, organ segmentation, and automated report generation.
Core Capabilities
Seven competencies across medical image analysis, from automated chest X-ray reading to visual question answering and radiology report generation.
Pathology Classification
Detect and classify 18 thoracic pathologies using DenseNet-121 and CheXagent with confidence scores for each finding.
Automated Radiology Reports
Generate structured radiology reports with imaging findings and diagnostic impressions using Maira-2, achieving 79.1% mF1-14 on MIMIC-CXR.
Organ Segmentation
Segment heart, lungs, and clavicles with pixel-level precision using MedSAM and PSPNet for quantitative anatomical analysis.
Visual Question Answering
Answer free-form medical questions about chest X-rays with 90.35% accuracy on the SLAKE benchmark via LLaVA-Med integration.
DICOM and Standard Formats
Process DICOM, JPEG, and PNG medical images with built-in visualization tools for clinical and research workflows.
Parallel Tool Execution
Run independent analysis tools concurrently to minimize wait times and deliver faster diagnostic insights.
Verified Performance on Medical Benchmarks
Results from peer-reviewed evaluations published at ICML 2025.
ChestAgentBench
63.1%
overall score (outperforms GPT-4o by 6.7 pts)
SLAKE VQA
90.35%
accuracy on visual question answering
Pathologies Detected
18
thoracic conditions identified
What Researchers and Clinicians Say
Feedback from professionals using this AI-powered radiology assistant in research and clinical settings.
Alex
Radiologist
The multi-tool orchestration approach solves a real problem — I can query a single system instead of running five separate models for a complex case.
Marie
Clinical Researcher
The automated report generation saves significant time during screening workflows. The structured output matches what we produce manually.
Jordan
Hospital IT Director
Being open-source with Apache-2.0 licensing was the deciding factor. We integrated it into our existing PACS workflow within a week.
Dr. Li
Thoracic Radiologist
The 18-pathology detection covers all the common findings we screen for. The confidence scores help prioritize which cases need immediate attention.
Dr. Wang
Medical AI Researcher
The modular architecture makes it straightforward to benchmark individual tools or add new ones. The LangGraph integration is well-designed.
Dr. Zhao
Department Head, Imaging
We tested it on 675 clinical cases from ChestAgentBench. The ability to handle multi-step reasoning queries sets it apart from single-model approaches.
Frequently Asked Questions
Common questions about this AI chest X-ray analysis platform, its capabilities, and how to get started.
What is MedRAX and how does it analyze chest X-rays?
MedRAX is an AI agent that combines 8 specialized chest X-ray analysis tools with a large language model. It uses a ReAct (Reasoning and Acting) loop to dynamically select the right tool for each query — whether that's pathology classification, organ segmentation, or report generation.
How accurate is AI chest X-ray analysis with this system?
The system achieves 63.1% on ChestAgentBench (outperforming GPT-4o by 6.7 points), 90.35% accuracy on SLAKE visual question answering, and 79.1% mF1-14 on MIMIC-CXR report generation. These results were independently verified and published at ICML 2025.
What chest X-ray diseases can this AI detect?
The pathology classification module detects 18 thoracic conditions including Pneumonia, Cardiomegaly, Atelectasis, Pleural Effusion, Pneumothorax, Consolidation, Edema, and Emphysema, among others. Each detection includes a confidence score.
Is this radiology AI tool open-source?
Yes. The full codebase is available on GitHub under the Apache-2.0 license with over 1,100 stars. You can deploy it in your own environment, modify the tools, or contribute to the project.
What deep learning models does this radiology AI use?
The framework integrates 8 specialized models: DenseNet-121 and CheXagent for pathology classification, MedSAM and PSPNet for organ segmentation, Maira-2 and SwinV2 for report generation, LLaVA-Med for visual question answering, and RoentGen for image generation.
Can this AI generate automated radiology reports?
Yes. The report generation module produces structured reports with imaging findings and diagnostic impressions. It achieves 79.1% mF1-14 on the MIMIC-CXR benchmark, making it competitive with specialized report generation models.
How does the organ segmentation feature work?
The system uses MedSAM and PSPNet to segment anatomical regions including the heart, left lung, right lung, and clavicles with pixel-level precision. This enables quantitative measurements like cardiothoracic ratio calculation.
What image formats does this CXR analysis tool support?
The platform processes DICOM, JPEG, PNG, and WebP medical images. DICOM support includes metadata extraction and proper windowing for diagnostic-quality display.
How does this system compare to GPT-4o for medical imaging?
On the ChestAgentBench evaluation (2,500 medical queries across 675 clinical cases), this system scores 63.1% compared to GPT-4o's 56.4% — a 6.7-point improvement. The advantage comes from specialized tool integration rather than relying on a single general-purpose model.
Can I integrate this AI radiology assistant into my clinical workflow?
Yes. The system is built on LangChain and LangGraph with a modular architecture. It provides a Gradio-based interface for direct use, plus an API for integration into existing PACS, EHR, or research pipelines. No retraining is needed to add new tools.
Try the AI Chest X-Ray Analysis — Free and Open-Source
Upload a chest X-ray and see pathology detection, organ segmentation, and report generation in action. Apache-2.0 licensed, no account required.