PII/PHI De-identification Application

A Flask web application for de-identifying Protected Health Information (PHI) and Personally Identifiable Information (PII) from clinical text using a fine-tuned ClinicalBERT model.

Features

• Automatic detection of PHI/PII entities (names, dates, locations, etc.)
• Surrogate data generation for consistent de-identification
• Web-based interface for easy text processing
• Support for staff vs. patient name classification
• Date shifting for temporal anonymization

Technology Stack

• Flask - Web framework
• PyTorch - Deep learning framework
• Transformers - Hugging Face library for BERT models
• ClinicalBERT - Fine-tuned model (obi/deid_bert_i2b2) for clinical NER

Local Setup

1. Clone the repository:

git clone https://github.com/rajellumen/pii-phi.git
cd pii-phi

2. Create a virtual environment:

python -m venv .venv
.venv\Scripts\activate  # Windows
# or
source .venv/bin/activate  # Linux/Mac

3. Install dependencies:

pip install -r requirements.txt

4. Set environment variables:

# Create a .env file or set environment variables
FLASK_SECRET_KEY=your-secret-key-here
DEID_API_KEY=your-api-key-here  # Optional, for API endpoint protection

5. Run the application:

python app.py

The application will be available at http://localhost:9000

Deployment on Amazon EC2

Prerequisites

1. AWS EC2 instance (Ubuntu recommended)
2. Python 3.10+ installed
3. Git installed

Deployment Steps

1. Connect to your EC2 instance:

ssh -i your-key.pem ubuntu@your-ec2-ip

2. Clone the repository:

git clone https://github.com/rajellumen/pii-phi.git
cd pii-phi

3. Set up Python virtual environment:

python3 -m venv venv
source venv/bin/activate

4. Install dependencies:

pip install --upgrade pip
pip install -r requirements.txt

5. Set environment variables:

export FLASK_SECRET_KEY="your-secret-key-here"
export DEID_API_KEY="your-api-key-here"  # Optional
export PORT=9000  # Or your preferred port

6. Run with Gunicorn (Production):

gunicorn -w 2 -b 0.0.0.0:9000 app:app

7. Or run as a systemd service (Recommended):
   • Create /etc/systemd/system/pii-phi.service:

   [Unit]
   Description=PII/PHI De-identification Service
   After=network.target
   
   [Service]
   User=ubuntu
   WorkingDirectory=/home/ubuntu/pii-phi
   Environment="PATH=/home/ubuntu/pii-phi/venv/bin"
   Environment="FLASK_SECRET_KEY=your-secret-key-here"
   Environment="DEID_API_KEY=your-api-key-here"
   ExecStart=/home/ubuntu/pii-phi/venv/bin/gunicorn -w 2 -b 0.0.0.0:9000 app:app
   
   [Install]
   WantedBy=multi-user.target

   • Enable and start:

   sudo systemctl enable pii-phi
   sudo systemctl start pii-phi

Important Notes for EC2 Deployment

• Security Groups: Ensure port 9000 (or your chosen port) is open in EC2 Security Groups
• Model Download: First run will download the BERT model (~500MB), which may take 5-10 minutes
• Memory Requirements: The model requires significant memory. Use at least 2GB RAM instance
• HTTPS: Use a reverse proxy (nginx) with SSL certificate for production
• Firewall: Configure ufw to allow your port:

  sudo ufw allow 9000/tcp

Environment Variables Reference

Variable	Required	Description
FLASK_SECRET_KEY	Yes	Secret key for Flask session management
DEID_API_KEY	No	API key for /deidentify endpoint protection
PORT	Auto	Port number (automatically set by Render)

Usage

1. Navigate to the web interface
2. Paste or type clinical text containing PHI/PII
3. Click "De-identify Text"
4. View the de-identified text with highlighted changes
5. Review the changes log showing all replacements

Model Information

• Model: obi/deid_bert_i2b2
• Purpose: Named Entity Recognition for clinical text
• Entity Types: PATIENT, STAFF, DATE, LOCATION, etc.

License

[Add your license here]

Author

Raj Anantharaman
Email: ranantharaman@ellumen.com