MediAgent — Autonomous Radiology Pipeline

MediAgent is an autonomous multi-agent medical imaging analysis system built entirely on AMD Instinct MI300X infrastructure. It addresses a critical global gap: over 2.3 billion people lack access to radiologists, and in regions like Kurdistan, patients can wait weeks for a single X-ray interpretation. MediAgent enables fast, local, and reliable analysis without cloud dependency. The system runs the Qwen3.6-27B multimodal model via vLLM on MI300X with 192GB HBM3 memory. This ensures zero API costs and guarantees that patient data never leaves the hospital, making it inherently privacy-preserving and compliant with standards like HIPAA. MediAgent operates through five specialized AI agents: The Intake Agent validates images, standardizes clinical input, extracts demographics, and detects safety flags. The Vision Agent performs structured radiological analysis, identifying findings with severity levels (NORMAL, INCIDENTAL, SIGNIFICANT, CRITICAL) and confidence scores. The Research Agent maps findings to a built-in knowledge base of conditions with ICD-10 codes, generating a ranked differential diagnosis with clinical reasoning. The Report Agent produces a formal radiology report aligned with American College of Radiology and National Institute for Health and Care Excellence standards. The Critic Agent reviews the report for consistency, flags uncertainty, enforces disclaimers, and assigns a QA score (0–100). The full pipeline completes in under 50 seconds via a clinical web interface with real-time tracking, severity color coding, and one-click PDF export. In testing, MediAgent accurately identified COVID-19 pneumonia patterns and normal scans. By keeping all computation local, it removes cost, latency, and privacy barriers—making scalable radiology support possible anywhere.