SENTINEL | Automated Medical Imaging Triage Framework SENTINEL is a simulated medical imaging triage system designed to automate the prioritization of radiological scans using both metadata analysis and Generative AI. It mimics a real-world PACS (Picture Archiving and Communication System) environment by accepting DICOM files, anonymizing patient data, and flagging critical cases for immediate review.
🚀 System Architecture The project is divided into three core components:
- Backend (FastAPI & DICOM Server) FastAPI Web Server: Handles authentication, image rendering, audit logging, and WebSocket communication for real-time dashboard updates.
DICOM SCP (Service Class Provider): A background server running on port 11112 that accepts incoming C-STORE requests using pynetdicom.
Triage Engine: A rule-based processing unit that anonymizes DICOM datasets and assigns urgency levels (Routine, Urgent, Critical) based on metadata keywords (e.g., "TRAUMA", "STROKE") and pixel intensity anomalies.
AI Integration: Utilizes the Gemini API to analyze DICOM pixel data for visual anomalies and tumor/foreign object detection.
- Frontend (Modern Web Interface) Triage Control Dashboard: A real-time interface built with HTML/CSS and Vanilla JS that displays the current study queue and system metrics.
Real-time Updates: Uses WebSockets to stream metrics (uptime, deliverability) and new study arrivals directly to the UI.
DICOM Viewer: A dedicated viewer for authorized personnel to render and inspect radiological images as PNGs.
- Simulation Tools Live Hospital Feed Simulator: A Python script (live_feed.py) that generates synthetic DICOM data with randomized metadata and pixel anomalies, then streams them to the Sentinel server at set intervals.
🛠 Tech Stack Backend: Python, FastAPI, Pydicom, Pynetdicom, NumPy, Pillow.
AI: Google Generative AI (Gemini 1.5 Flash).
Frontend: HTML5, CSS3, JavaScript, WebSockets, Three.js/Vanta.js (for aesthetics).
📂 Core Files backend/main.py: Main entry point for the FastAPI server and DICOM SCP.
backend/triage_engine.py: Logic for anonymization and rule-based triage.
backend/dicom_to_gemini.py: Pipeline for converting DICOM to image bytes and querying Gemini.
frontend/index.html: The main dashboard layout.
test_client/live_feed.py: Simulator for generating and sending DICOM traffic.
🔒 Security & Privacy Anonymization: Every incoming DICOM file is stripped of identifying information (Name, ID, BirthDate) and assigned a unique anonymous UID before storage.
Audit Logging: Every sensitive action—including logins, logouts, and image rendering—is logged to a secure terminal audit trail.
Access Control: The system uses a predefined set of Valid Access IDs (e.g., R-001 for Dr. Rishi) to authorize API requests and image access.
⚙️ Setup & Installation Install Dependencies:
Bash pip install -r backend/requirements.txt Configure API Key: Update the api_key in backend/main.py and backend/dicom_to_gemini.py with your Gemini API Key.
Run the Backend:
Bash uvicorn main:app --reload Start the Simulator:
Bash python test_client/live_feed.py Access the Dashboard: Open frontend/login.html in your browser and log in with a valid ID (e.g., R-001).