MedLens: A Multi-Agent AI Medical Imaging System for Intelligent Triage and Report Generation of Chest X-Rays

MedLens is an award-winning project from Catalyst 2k26, a multi-agent AI system designed specifically for chest X-ray triage and report generation. The system integrates computer vision, Retrieval-Augmented Generation (RAG), Explainable AI (XAI), and validation processes to provide intelligent support for clinical diagnosis. Maintained by pscommits, the project was released on June 2, 2026, and its source code is available on GitHub (link: https://github.com/pscommits/MedLens).

Medical imaging diagnosis is a critical part of healthcare, but the global shortage of radiologists leads to delayed reports, affecting treatment timing. Traditional CAD systems only focus on image classification and cannot generate standardized reports; medical AI needs to meet high reliability, explainability, and adaptability requirements, which are the main obstacles to its implementation.

MedLens adopts a multi-agent architecture, simulating the division of labor and collaboration mode in hospitals, breaking down complex tasks into subtasks handled by specialized agents. Core design principles include clinical implementation (supporting DICOM format and HL7 protocol). The technology stack covers: computer vision module (image understanding and anomaly detection), RAG (real-time retrieval of medical knowledge), XAI (visualization of decision-making basis), and validation processes (multi-layer quality assurance).

The MedLens workflow consists of five steps: 1. The triage agent assesses the urgency level; 2. The image analysis agent identifies abnormal signs and generates structured findings; 3. The knowledge retrieval agent synchronously retrieves relevant medical knowledge; 4. The report generation agent integrates information to produce a standardized report; 5. The validation agent reviews the report's logic, accuracy, and format, and modifies it if it fails to pass.

MedLens can improve the accessibility of medical quality (providing virtual expert support for resource-poor areas), help large hospitals share screening and report drafting work, reduce the rate of missed diagnoses and misdiagnoses, and shorten report time. Its explainable design meets medical regulatory requirements and is expected to obtain clinical application approval.

MedLens demonstrates the potential of multi-agent architecture in vertical fields, emphasizing the necessity of RAG and XAI in high-risk scenarios (non-black-box systems). It provides a reference paradigm for AI developers: they need to combine industry knowledge, interdisciplinary cooperation, and rigorous engineering practices to transform cutting-edge technology into practical solutions.

In the future, MedLens can evolve towards multi-modal fusion (integrating X-rays, CT, MRI, etc.) and personalized diagnosis (combining patient medical history and genetic information). With the improvement of large model capabilities and the accumulation of medical data, AI applications in the medical imaging field will become more in-depth, and MedLens' experience can provide a reference for subsequent projects.