Disentangled Representational Learning of Single Lead Electrocardiogram Signals using Variational Autoencoder

This work focuses on clustering 1-lead electrocardiogram (ECG) heartbeats using beta total correlation variational autoencoder ($\beta$-TCVAE). The objective is to detect irregular morphologies in ECG signals, which can serve as indicators of cardiac anomalies.

Installation and Setup

To get started with the project, follow these steps:

1. Make sure you have Python version 3.10 installed.
2. Create a virtual environment
3. Install the required libraries by running the following command in the project directory. Some requirements might need adjustemnt depending on your hardware and OS:

conda create -n ecg python=3.10
conda activate ecg
pip install -r requirements.txt

Data Preparation

The raw ECG data is available in a remote repository and needs to be downloaded and built. Therefore, perform the following steps:

1. Clone the ECG-TFDS repository:

git clone https://github.com/CardioKit/ECG-TFDS

2. Install the requirements for ECG-TFDS:

pip install -r ./ECG-TFDS/requirements.txt

3. Change to the ECG-TFDS source directory (e.g., Zheng's dataset):

cd ./ECG-TFDS/src/zheng

4. Build the dataset:

tfds build --register_checksums

Running the Code

Execute the main file to run the code:

python main.py 

The main file requires a configuration file for parameterization:

options:
  -h, --help            show this help message and exit
  -p, --path_config     location of the params file (default: ./params.yml)

Evaluation

The results of the runs can be analyzed with the jupyter notebook:

./analysis/article.ipynb

How to cite?

If you want to either use code or refer to results, please cite the following article: (To be determined)

@article{kapsecker2024disentangled,
  title={Disentangled Representational Learning of Single Lead Electrocardiogram Signals using Variational Autoencoder},
  author={Kapsecker, Maximilian and Möller, Matthias C and Jonas, Stephan M},
  journal={TBD},
  volume={TBD},
  number={TBD},
  pages={TBD},
  year={TBD},
  publisher={TBD}
}