CORE-BEHRT: Carefully Optimized and Rigorously Evaluated BEHRT

Overview

CORE-BEHRT is an advanced framework for analyzing Electronic Health Records (EHR) using a BERT-like model optimized for healthcare data. This project aims to provide a robust, reproducible, and state-of-the-art solution for EHR data analysis.

Features

• Efficient data preprocessing pipeline
• Customizable BEHRT model architecture
• Comprehensive training and evaluation scripts
• Cross-validation support for robust model assessment
• Visualization tools for result interpretation

Prerequisites

Ensure you have the following dependencies installed:

• Python 3.7+
• PyTorch 1.7+
• transformers 4.0+
• NumPy
• Pandas
• scikit-learn
• tqdm
• matplotlib
• pyarrow (for Parquet file support)

Installation

1. Clone the repository:

   git clone https://github.com/your-username/core-behrt.git
   cd core-behrt

2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

3. Install the package:

   pip install -e .

Usage

Follow these steps to preprocess your data, train the model, and evaluate the results:

1. Data Preparation:

   python -m ehr2vec.scripts.main_create_data

2. Model Pre-training:

   python -m ehr2vec.scripts.main_pretrain

3. Prepare Fine-tuning Data:

   python -m ehr2vec.scripts.main_create_outcomes

4. Model Fine-tuning:

   python -m ehr2vec.scripts.main_finetune_cv

To use a custom configuration, pass the path to your config file: