🏥 Federated Multimodal Leukemia Diagnosis System

A production-ready Clinical Decision Support System (CDSS) utilizing Federated Learning, Dual-Input Multimodal Neural Networks (Image + Tabular Data), and Explainable AI (XAI) for the secure and interpretable detection of acute leukemia.

---

⚡ Fastest Start (2 Steps)

Terminal 1 - Backend (FastAPI):

cd backend 
python -m venv venv
# Windows: venv\Scripts\activate | macOS/Linux: source venv/bin/activate
pip install -r requirements.txt
pip install opencv-python-headless scikit-learn  # Required for XAI and Scaler
python main.py

Terminal 2 - Frontend (Next.js):

cd frontend 
npm install
npm install lucide-react recharts axios react-hot-toast # Ensure new UI dependencies are installed
npm run dev

Then open: http://localhost:3000

---

✨ Enterprise Features

🎯 Federated Learning & Model Selection

• Decentralized AI: Employs a Federated Averaging (FedAvg) strategy to train models across institutions without pooling raw patient data.
• Dynamic Model Switching: Users can toggle between the Global Aggregated Model (99.2% accuracy) and local institutional models (Site Alpha / Site Beta) directly from the UI.
• Differential Privacy: Defends against model inversion attacks using local gradient clipping and additive Gaussian noise.

🧠 Explainable AI (XAI) & Safety

• Visual Interpretability (Grad-CAM): Generates heatmaps over blood smears to prove the neural network is focusing on morphological anomalies, not background artifacts.
• Tabular Feature Attribution: Calculates partial derivatives to show exactly which clinical blood counts (e.g., high WBC, low platelets) drove the diagnosis.
• Uncertainty Quantification: Utilizes Monte Carlo (MC) Dropout to run multiple inference passes, calculating prediction variance to warn doctors of Out-of-Distribution (OOD) data.

🧬 Multimodal Late-Fusion Architecture

• Dual-Input Analysis: Combines 224x224 blood smear microscopy images (CNN branch) with 9 clinical laboratory values (MLP branch).
• Automated Data Processing: Real-time image normalization, resizing, and tabular standard scaling (joblib).
• One-Click Demo Profiles: Pre-loaded clinical profiles (Healthy/Leukemia) for fast system validation.

🎨 Professional UI & Interactive Documentation

• Glassmorphism Design: Sleek, modern dark-themed interface utilizing Tailwind CSS and Lucide React icons.
• Native Data Visualization: Built-in documentation page (/documentation) featuring interactive recharts graphs proving FL convergence, non-IID data handling, and local vs. global performance.

---

🏗️ System Architecture

Browser (localhost:3000)
         ↓ (FormData: Image + 9 Features + Model Choice)
    Frontend (Next.js + React)
    - Dynamic Model Selector
    - Demo Data Autofill
    - XAI Visualization Dashboard
         ↓ HTTP POST /predict
    Backend (FastAPI - localhost:8000)
    - Image & Tabular Preprocessing
    - Grad-CAM & Gradient Math Engine
         ↓
    ML Model (TensorFlow/Keras)
    - Late-Fusion Multimodal Network
    - Binary Focal Crossentropy (Handles Class Imbalance)

---

📁 Project Structure

leukemia-diagnosis/
├── backend/                    # FastAPI application & ML Inference
│   ├── main.py                 # Core server logic & XAI generators
│   └── requirements.txt        
├── frontend/                   # Next.js application
│   ├── app/page.tsx            # Main Diagnostic Dashboard
│   ├── app/documentation/      # Architecture & FL Results Page
│   └── tailwind.config.ts      
├── models/                     # Federated ML assets
    ├── base models/            # Local Institutional Models
    │   ├── local_model_alpha.keras
    │   └── local_model_beta.keras
    ├── global models/          # Aggregated Models
    │   └── global_model.keras  
    └── scaler/                 
        └── scaler_global.joblib # Scikit-learn standardization rules

---

📊 API Reference

Base URL: http://localhost:8000

GET /health - Health check & asset verification

{ 
  "status": "healthy", 
  "models_loaded": {"global": true, "alpha": true, "beta": true}, 
  "scaler_loaded": true 
}

POST /predict - Multimodal Inference & XAI Generation

• Input (FormData): file (Image), WBC_count ... Uric_acid (Floats), model_type (String: "global"|"alpha"|"beta")
• Output:

{
  "classification": "Leukemia",
  "confidence": 0.985,
  "model_used": "global",
  "explanation_image": "<base64_gradcam_string>",
  "feature_importance": {"WBC_count": 0.45, "Platelet_count": -0.12...}
}

---

⚠️ Clinical Disclaimer

This system is a Clinical Decision Support System (CDSS) prototype built for research and educational purposes under a Federated Learning framework.

• It is NOT a substitute for professional medical diagnosis.
• Results should NOT be used for primary clinical decisions.
• Always consult a qualified hematologist or oncologist.