midcine

Zero-code radiology AI assistant that runs inside your hospital firewall — DICOM in, structured reports out.

---

🩻 Live demo (open in browser — no install)

► https://ame.tail19ddab.ts.net:8445/reports ← click any case in the worklist

► Open a sample study directly (MR spine)

The demo runs on a private machine over Tailscale Funnel — availability follows the host's uptime. All patient names, IDs, and free-text clinical fields are synthetic; DICOM pixel data is anonymized. A yellow ⚠ DEMO / FICTIONAL DATA banner is visible on every page.

What to try on the demo:

• Open /reports → pick a case → 2D viewer with window/level, zoom, pan, length + angle tools
• Open /viewer/<uid> → NiiVue MPR / 3D volume rendering
• Voice dictate + AI report drafting are wired (may take 15–90 s per request; hosted LLM)
• Upload your own anonymized DICOM at /upload (files stay on host, never re-shared)

---

Who reads this?

Hospital IT administrators who need to deploy midcine in under 5 minutes on existing Docker infrastructure.

---

TL;DR

1. Clone → docker compose up -d → paste your Naraya API key.
2. Verify with curl http://localhost:8210/health (FastAPI) and curl http://localhost:3000 (Next.js UI).
3. Optional: enable PACS, Ollama fallback, or TotalSegmentator via docker-compose.premium.yml.

---

Prerequisites

• Docker Engine ≥ 24.0 (v20.10+ minimum)
• Docker Compose ≥ 2.23 (v2.0+ minimum)
• Linux host (tested on Ubuntu 22.04 LTS)
• 8GB RAM minimum (16GB recommended for premium services)
• Outbound HTTPS (port 443) to Naraya free tier (api.bynara.ai)

---

3-step deploy

1. Clone the repo

git clone https://github.com/midcine/midcine.git && cd midcine

2. Set the required environment variable

echo "NARAYA_API_KEY=your_free_tier_key_here" > .env

3. Start the stack

docker compose up -d

---

Verify

FastAPI (mcp-bridge) health check

curl -s http://localhost:8210/health | jq
# Expected: {"status":"ok","version":"0.4.2"}

Next.js UI

curl -s http://localhost:3000 | grep -q "midcine" && echo "UI up" || echo "UI down"

---

Optional upgrades

Enable premium services by merging the premium compose file:

# Option 1: Override file
cp docker-compose.premium.yml docker-compose.override.yml
docker compose up -d

# Option 2: Direct compose
docker compose -f docker-compose.yml -f docker-compose.premium.yml up -d

Service	Purpose	Env var needed
Orthanc PACS	Local DICOM store	None
Ollama fallback	Local LLM fallback	OLLAMA_URL=http://host.docker.internal:11434 or http://ollama:11434
TotalSegmentator	Whole-body segmentation	None
backup-cron	Nightly backups	BACKUP_S3_BUCKET=your-bucket

---

Common issues

1. Port 8210 already in use → Stop conflicting service (sudo lsof -i :8210) or edit docker-compose.yml line 12 to expose a different port.

2. NARAYA_API_KEY rejected → Sign up for a free tier key at https://bynara.ai and ensure the key is pasted without whitespace or quotes in .env.

3. Next.js UI stuck on loading spinner → Check FastAPI logs (docker compose logs mcp-bridge) for 502 errors; ensure NARAYA_API_KEY is set in .env.

4. Ollama connection refused → Ensure Ollama is running on host and OLLAMA_URL points to correct address.

---

Next steps

• User guide for radiologists
• Configure PACS connection
• Customize LLM prompts
• API reference for integrators
• Contributing & local development