Drug Dosage Explanation and Medical Assistant

An intelligent healthcare application that explains medication dosages based on prescription information and provides medical assistance through a dual-mode chat interface.

Features

• Prescription Image Processing: Extract text from prescription images using OCR
• Dosage Prediction: Machine learning model to predict appropriate dosages
• Explainable AI: Understand why a particular dosage was recommended
• User-friendly Interface: Simple web interface built with Streamlit
• Multiple Input Methods: Upload prescription images or enter details manually
• Dual-Mode Medical Assistant:
  • Independent Mode: Ask general medical questions without requiring a prescription
  • Enhanced Mode: Get personalized answers based on your prescription data

Prerequisites

• Python 3.8+
• Tesseract OCR (for image processing)
• Google Gemini API key (optional, for enhanced text processing)

Installation

1. Clone the repository:

   git clone <repository-url>
   cd drug-dosage-chatbot

2. Install Tesseract OCR:

   • Windows: Download and install from UB Mannheim
   • macOS: brew install tesseract
   • Linux: sudo apt-get install tesseract-ocr

3. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

4. Install the required packages:

   pip install -r requirements.txt

5. (Optional) Set up Google Gemini API:

   • Get an API key from Google AI Studio
   • Create a .env file in the project root and add:

     GEMINI_API_KEY=your_api_key_here
     

Usage

Quick Start

The easiest way to run the application is using the launcher script:

python run_app.py

This script will:

1. Check if your Gemini API key is set
2. Install required dependencies
3. Start the Streamlit application

Manual Setup

1. Set up Google Gemini API:

   • Get an API key from Google AI Studio
   • Set it as an environment variable:

     # Windows (PowerShell)
     $env:GEMINI_API_KEY="your-api-key-here"
     
     # Windows (Command Prompt)
     set GEMINI_API_KEY=your-api-key-here
     
     # Linux/Mac
     export GEMINI_API_KEY=your-api-key-here

2. Generate Synthetic Data (Optional):

   python data_generator.py

3. Train the Model (Optional):

   python train_model.py

4. Run the Web Application:

   streamlit run app.py

How to Use

1. Upload Prescription:

   • Click on the "Upload Prescription" tab
   • Upload an image of your prescription
   • Click "Analyze Prescription" to process the image

2. Or Enter Details Manually:

   • Click on the "Enter Details Manually" tab
   • Fill in the required information
   • Click "Get Dosage Recommendation"

3. View Results:

   • See the recommended dosage and frequency
   • Expand the explanation to understand how the recommendation was made
   • Review important safety information

4. Chat with Medical Assistant:

   • Click on "Chat with Medical Assistant"
   • Ask questions about your prescription if one is loaded
   • Ask general medical questions even without a prescription

Project Structure

• app.py: Streamlit web application
• enhanced_chatbot.py: Improved chatbot with independent Gemini interface
• independent_gemini.py: Standalone Gemini interface for general medical queries
• chatbot.py: Original chatbot logic and prediction pipeline
• gemini_chat.py: Original Gemini chat implementation
• data_generator.py: Script to generate synthetic prescription data
• train_model.py: Script to train the dosage prediction model
• xai_module.py: Explainable AI functionality using SHAP
• xai_module_simple.py: Simplified XAI when SHAP is not available
• ocr_parser.py: OCR and text processing for prescription images
• pages/chat.py: Streamlit chat interface page
• run_app.py: Launcher script with dependency checks
• models/: Directory for trained models
• data/: Directory for datasets (not included in version control)

Architecture

Dual-Mode Medical Assistant

The application provides two modes of operation:

1. Independent Mode:

   • Can answer general medical questions without requiring a prescription
   • Uses Google's Gemini API to generate responses
   • Works even when no prescription data is available

2. Enhanced Mode:

   • Provides personalized answers based on prescription data
   • Incorporates patient-specific information in responses
   • Leverages the dosage prediction model for more accurate advice

Graceful Degradation

The system is designed to handle missing components gracefully:

• If SHAP is not available, it falls back to a simplified explainer
• If no prescription data is available, it still provides general medical advice
• If OCR fails on a prescription image, it allows manual entry of data

Customization

Adding New Medications

To add support for additional medications:

1. Update the medicine_mapping in ocr_parser.py
2. Add the medication to the valid_diseases dictionary in chatbot.py
3. Add appropriate safety notes in xai_module.py
4. Retrain the model with updated data

Modifying the Model

To use a different machine learning model:

1. Update the model pipeline in train_model.py
2. Modify the prediction logic in chatbot.py
3. Update the explanation code in xai_module.py if needed

Limitations

• This is a prototype and not intended for clinical use
• The model is trained on synthetic data and may not be accurate for all cases
• Always consult with a healthcare professional for medical advice

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Built with Streamlit, scikit-learn, and Google's Gemini API
• Icons by Font Awesome
• Color scheme inspired by Material Design