This project is an AI-powered clinical decision support system that assists healthcare professionals by analyzing patient reports and providing preliminary diagnostic insights. It leverages a Retrieval-Augmented Generation (RAG) architecture to deliver comprehensive and context-aware analysis.

The system takes a patient's medical report (in PDF format) as input, extracts relevant clinical data, and then compares it against a vast database of existing medical cases. By identifying similarities and patterns, it generates a detailed report that includes:

• Potential diseases or health concerns
• Recommended precautions and lifestyle changes
• Suggestions for FDA-approved medications and first-aid
• Actionable insights for doctors, supported by data from similar cases
• A clear list of the data points that informed the conclusion

This tool is designed to augment the expertise of medical professionals, not to replace it. The AI-generated analysis should always be reviewed and validated by a qualified doctor.

• Multi-Disease Analysis: The system is trained on datasets for various conditions, including diabetes, kidney stones, heart disease, and anemia.
• RAG Architecture: It uses a Retrieval-Augmented Generation (RAG) model to ground its analysis in real-world data, improving accuracy and relevance.
• Vector Similarity Search: Employs Qdrant and sentence-transformer embeddings to efficiently find similar patient cases from the knowledge base.
• Web-Enhanced Insights: Augments its analysis with real-time information from the web, ensuring the recommendations are current.
• Comprehensive Reporting: Generates detailed, multi-section reports to support clinical decision-making.

1. Data Preprocessing: Medical datasets (CSV files) are cleaned, transformed, and stored as a collection of documents.
2. Embedding: The processed documents are converted into vector embeddings using a sentence-transformer model and stored in a Qdrant vector database.
3. Patient Report Analysis: When a new patient report (PDF) is provided, the system extracts the clinical text.
4. Similarity Search: The extracted text is used to query the Qdrant database, retrieving the most similar patient cases.
5. Web Search: An AI-generated query is used to search the web for additional, relevant medical information.
6. Report Generation: A large language model (Llama 3.1) synthesizes the information from the patient report, similar cases, and web search results to generate a final, comprehensive analysis.

Follow these steps to set up and run the project locally.

• Python 3.8+
• Pip for package management
• A Hugging Face API token
• A Qdrant account (for cloud storage) or a local Qdrant instance

1. Clone the repository:

   git clone https://github.com/your-username/Healthcare-Assistant.git
   cd Healthcare-Assistant

2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate

3. Install the required packages:

   pip install -r requirements.txt

4. Set up your environment variables: Create a .env file in the root directory and add your API keys:

   HUGGINGFACE_API_TOKEN="your_huggingface_api_token"
   QDRANT_API_KEY="your_qdrant_api_key"
   

The data directory is not tracked by Git. You will need to create it and populate it with the necessary raw data files.

1. Create the directory structure:

   mkdir -p data/raw
   

2. Add the raw data files:

   Place your raw CSV data files in the data/raw/ directory. The project is configured to use the following files:

   • diabetes_classification.csv
   • kidney_stone_dataset.csv
   • heart_disease.csv
   • anemia.csv
   • thyroidDf.csv

   Your data directory should look like this:

   data/
   └── raw/
       ├── diabetes_classification.csv
       ├── kidney_stone_dataset.csv
       ├── heart_disease.csv
       ├── anemia.csv
       └── thyroidDf.csv
   

The following datasets are used in this project:

• Kidney Stone Prediction
• Heart Disease Prediction
• Malaria Detection
• Anemia Prediction
• Thyroid Disease Prediction

1. Preprocess the data: Run the preprocessing script to prepare the datasets.

   python src/pre-processing/preprocess.py

2. Create the embeddings: Generate embeddings from the preprocessed data and store them in your Qdrant database.

   python src/embeddings/create_embeddings.py

3. Run the analysis: Place your patient report PDF in the data/raw/ directory (e.g., test_report_2.pdf) and run the retrieval script.

   python src/embeddings/retrieve_embeddings.py

   The system will output the final analysis to the console.