AI-powered MRI DICOM scan analysis using Claude.

Screenshots

• DICOM file loading - from local folder or drag & drop
• AI-assisted series selection - describe what you want to examine ("spine", "brain") and AI selects relevant series
• Multi-series selection - manually select multiple series
• Auto-filtering - localizer/scout images are automatically filtered
• Slice navigation - mouse wheel, arrow keys, slider
• Analysis modes:
  • Overview (~10 slices) - quick overview
  • Current - current slice only
  • Selected - manually selected slices
  • All - all loaded slices
• AI annotations - markers on images where AI found findings
• Contextual chat - continue conversation with AI about the scans

mri/
├── backend/                 # Python FastAPI
│   ├── main.py             # API endpoints
│   ├── dicom_utils.py      # DICOM parsing
│   ├── claude_client.py    # Claude API
│   └── requirements.txt
├── frontend/               # React + TypeScript + Vite
│   ├── src/
│   │   ├── App.tsx
│   │   ├── components/
│   │   │   ├── AuthScreen.tsx      # Token input
│   │   │   ├── DicomViewer.tsx     # DICOM viewer
│   │   │   ├── ChatPanel.tsx       # AI chat
│   │   │   └── UploadZone.tsx      # File upload
│   │   ├── hooks/
│   │   │   ├── useApi.ts
│   │   │   └── useStore.ts
│   │   └── types/
│   └── package.json
└── uploads/                # Temporary storage

• Python 3.11+
• Node.js 18+
• Claude API key (sk-ant-api03-*)

cd backend

# Create virtual environment
python3 -m venv venv

# Activate (macOS/Linux)
source venv/bin/activate

# Activate (Windows)
# venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

cd frontend

# Install dependencies
npm install

./start.sh

make install  # First time only
make dev      # Start both servers

Terminal 1 - Backend:

cd backend
source venv/bin/activate
uvicorn main:app --reload --port 8000

Terminal 2 - Frontend:

cd frontend
npm run dev

Open http://localhost:5173

1. Enter API token - Claude API key (starts with sk-ant-api03-)

2. Load DICOM files:

   • Click "Load from local folder" and enter path to DICOM folder
   • Or drag & drop files into the upload zone

3. Select series:

   • AI selection: enter what you want to examine (e.g., "spine and lower back")
   • Manual selection: click on series checkboxes
   • Load All: load all non-localizer series

4. Navigation:

   • Mouse wheel / arrow keys - switch slices
   • Space - select slice for analysis
   • Slider - quick navigation

5. Analysis:

   • Select mode: Overview / Current / Selected / All
   • Enter question or use Quick Actions
   • AI will show findings with markers on images

• FastAPI - async API server
• pydicom - DICOM parsing
• Pillow + numpy - image processing
• anthropic - Claude API SDK

• React 18 + TypeScript
• Vite - build tool
• TailwindCSS - styling
• Zustand - state management
• react-dropzone - file upload
• react-markdown - AI response rendering

Uses Claude Sonnet 4 (claude-sonnet-4-20250514).

The system prompt is configured for:

• Structural anomaly analysis
• Unusual signal detection
• Mass effect identification
• Atrophy detection
• Returning finding coordinates for marker display

• API token stored only in sessionStorage (not localStorage)
• Token passed in header with each request
• Files deleted after session
• Backend does not store tokens

If pyenv Python has SSL issues:

# Use Homebrew Python
/opt/homebrew/bin/python3 -m venv venv
./venv/bin/pip install -r requirements.txt

Frontend must run on localhost:5173 (configured in CORS).

Make sure token starts with sk-ant-api03- (not sk-ant-sid or others).

AI analysis is for educational and informational purposes only. This is NOT a medical diagnosis. Always consult a qualified radiologist for professional interpretation.