🧠 MirAI - Early Alzheimer's Screening System

AI-powered 3-tier screening that catches risk before symptoms appear.

🌐 Live Demo

Static Demo: https://varghese778.github.io/codeversity-MirAI/

⚠️ The live demo uses client-side mock calculations. For real ML model predictions, follow the local setup instructions below.

---

🚀 Quick Start (Local with Real ML Models)

Prerequisites

• Python 3.8+
• pip

Step 1: Clone the Repository

git clone https://github.com/Varghese778/codeversity-MirAI.git
cd codeversity-MirAI

Step 2: Install Dependencies

cd backend
pip install -r requirements.txt

Step 3: Train the ML Models

jupyter notebook model.ipynb

Run all cells in the notebook. This will:

• Load ADNI data from data/ folder
• Train 3-stage cascade models (Clinical → Genetic → Biomarker)
• Save model artifacts to models/ folder

Step 4: Run the Flask Server

cd ..
python app.py

Step 5: Open in Browser

Navigate to: http://localhost:5000

---

📁 Project Structure

codeversity-MirAI/
├── index.html              # Homepage
├── assessment.html         # Risk assessment wizard
├── results.html            # Results visualization
├── app.py                  # Flask backend server
├── assets/                 # CSS, JS, images, vendor libs
│   ├── css/main.css        # MirAI color scheme
│   └── vendor/             # Bootstrap, AOS, etc.
├── backend/
│   ├── model.ipynb         # Jupyter notebook for training
│   ├── mirai_inference.py  # Inference script
│   ├── requirements.txt    # Python dependencies
│   ├── data/               # ADNI CSV files
│   │   ├── ADNIMERGE_01Feb2026.csv
│   │   ├── APOERES_01Feb2026.csv
│   │   ├── UPENN_PLASMA_*.csv
│   │   └── ...
│   └── models/             # Trained model artifacts (generated)
│       ├── stage1_pipeline.pkl
│       ├── stage2_pipeline.pkl
│       ├── stage3_pipeline.pkl
│       └── *.json
└── README.md

---

🔬 The 3-Stage Pipeline

Stage	Name	Features	Purpose
1	Clinical Screening	Age, Gender, Education, FAQ, ECog	Non-invasive initial triage
2	Genetic Stratification	Stage1_Prob + APOE ε4	Risk refinement using genetics
3	Biomarker Confirmation	Stage2_Prob + pTau-217, Aβ42, NfL	Blood-based pathology detection

Model Performance (on ADNI data)

• Stage 1: AUC ~0.87
• Stage 2: AUC ~0.88
• Stage 3: AUC ~0.93

---

🛠️ API Reference

POST /api/predict

Request Body:

{
  "age": 72,
  "gender": "Female",
  "education": 16,
  "faq": 8,
  "ecogMem": 2.5,
  "ecogTotal": 2.0,
  "genotype": "3/4",
  "ptau217": 0.45,
  "ab42": 15.2,
  "ab40": 180.5,
  "nfl": 22.0
}

Response:

{
  "success": true,
  "prediction": {
    "final_risk_score": 0.62,
    "final_risk_category": "Elevated",
    "stage1_risk": "Elevated",
    "stage2_risk": "Elevated",
    "stage3_risk": "Elevated",
    "top_factors": [
      "FAQ Score: 8",
      "APOE4 Count: 1",
      "pTau-217: 0.45"
    ]
  },
  "model_type": "real"
}

---

👥 Team

Team Break&Build

• 📞 +91 9531975283
• 🔗 GitHub: @Varghese778

---

📚 Research Foundation

MirAI's methodology is grounded in established Alzheimer's research:

• Marshal Folstein - MMSE Developer
• Allen D. Roses - APOE ε4 Pioneer
• Oskar Hansson - pTau-217 Researcher

Data source: Alzheimer's Disease Neuroimaging Initiative (ADNI)

---

⚠️ Disclaimer

This is a screening tool, not a diagnostic. Results indicate relative risk based on inputs. Always consult a qualified healthcare provider for clinical evaluation.

---

📄 License

MIT License - Feel free to use and modify for research and educational purposes.