A comprehensive, research-grade ophthalmology platform integrating Deep Learning (Ensemble Models), RAG-powered AI Assistance, and Role-Specific Clinical Dashboards.
- Dual-Model Ensemble: Combines AlexNet and ResNet50 for high-accuracy classification of 6 retinal conditions (Cataract, Diabetic Retinopathy, Glaucoma, etc.).
- Explainable AI (Grad-CAM): Generates heatmaps to highlight exactly where the AI "sees" disease markers, enhancing clinical trust.
- Uncertainty Quantification: AI provides confidence scores to ensure safe screening.
- Multi-Agent Orchestration: Powered by Llama-3.1 and a specialized routing layer that delegates tasks to Diagnostician, Researcher, and Risk Analyst agents.
- C3-RAG Engine: Clinical Context-Constrained Retrieval Augmented Generation ensures zero-hallucination medical grounding against AAO guidelines.
- Action-Aware: Agents can modify doctor availability, view patient histories, and trigger real-time UI updates via natural language.
- Patient Dashboard: View AI reports, track appointment history, and book lab tests.
- Doctor Portal: Manage schedules, accept/reject appointments, and review detailed diagnostic findings with AI heatmaps.
- Lab Technician Interface: High-throughput retinal scan processing and verification.
To align with clinical standards, we have implemented five advanced research modules:
- Clinical Metrics Dashboard: Professional-grade evaluation showing Accuracy, Precision, Recall, F1, and interactive ROC Curves/Confusion Matrices for the ensemble.
- Multi-Modal AI Integration: High-fidelity diagnostics combining Fundus Image embeddings with OCR-extracted clinical text and doctor notes via a weighted evidence fusion engine.
- Counterfactual Explainability: A causal "What-If" simulation tool that allows clinicians to mask identified lesions (using Grad-CAM) and observe the AI's re-prediction, providing proof of diagnostic focus.
- Disease Progression Trajectory: Predictive analysis of patient history using Slope-based risk calculation to forecast 90-day disease progression.
- Multi-Agent Clinical Co-Pilot: Specialized assistant personas (Diagnostician, Researcher, Risk Analyst) with automated routing to provide role-specific medical context.
- Backend: Python (Flask), SQLAlchemy, SQLite
- AI/ML: PyTorch (AlexNet, ResNet50), Scipy
- Frontend: Next.js, Tailwind CSS, Framer Motion
- Knowledge Base: Groq/HuggingFace API, Custom RAG Implementation
- Remote Access: DuckDNS & ngrok integration
# Clone the repository
git clone https://github.com/SK4LEGENDS/Smart_Eye.git
cd Smart_Eye
# Create .env file (see Environment Variables section)
# Install dependencies
pip install -r requirements.txt
# Run the server
python app.pycd frontend
npm install
npm run devTo enable AI features, create a .env file in the root with:
HF_TOKEN: Your HuggingFace API Token.DUCKDNS_TOKEN: For permanent external access.SECRET_KEY: Your Flask secret key.
Smart Eye Care is developed with passion by:
- Jayaharini
- Kailash
- Jerlin John
This system is designed for screening and "Clinical Decision Support." It is intended for research and educational purposes and should not replace professional medical diagnosis.