Learning Generic Dissimilarity Representations with Large ECG Model for CVDs Detection

This repository contains the implementation for the paper "Learning Generic Dissimilarity Representations with Large ECG Model for CVDs Detection". The code includes pretraining scripts for tokenizers and a complete self-supervised learning process using the Mamba2 model. The generated representations can be evaluated on various downstream tasks for cardiovascular disease (CVD) detection.

Setup

1. Create a Virtual Environment

Before proceeding, create a Python virtual environment to manage dependencies:

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

2. Install Dependencies

Install the required dependencies listed in the requirements.txt file:

pip install -r requirements.txt

Data Preparation

All datasets should be pre-converted into HDF5 format. The organization of the data is as follows:

• Training Data: Store in the Datasets folder.
• Validation Data: Store in the Validation folder.
• Log Files: Store in the Logs folder.
• Parameter Files: Store in the Parameters folder.

Each dataset should be placed in a separate subfolder within the respective folders. All HDF5 files belonging to a specific dataset should be placed at the same level within its respective subfolder.

Pretraining

The pretraining phase consists of three scripts:

1. pretrainCrossTokenMulty.py: This script pretrains the AssocTokenizer to prevent gradient explosion during the self-supervised learning phase.

2. pretrainDeviatTokenMulty.py: This script pretrains the DeviaTokenizer for the same purpose.

3. pretrainECGLMMulty.py: This script performs the complete self-supervised learning using the Mamba2 model.

Performance Evaluation

After the pretraining and self-supervised learning phases, the generated files can be evaluated using the scripts in the ValExps folder. These scripts test the performance of the learned representations on various downstream classifiers for CVD detection.

Citation

If you find this repository useful in your research, please consider citing our paper:

@article{Haibo2024learning, title={Learning Generic Dissimilarity Representations with Large ECG Model for CVDs Detection}, author={Haibo Zheng}, journal={Journal Name}, year={2024}, }