DeepECG

ECG classification programs based on ML/DL methods. There are two datasets:

• training2017.zip file contains one electrode voltage measurements taken as the difference between RA and LA electrodes with no ground. It is taken from The 2017 PhysioNet/CinC Challenge.
• MIT-BH.zip file contains two electrode voltage measurements: MLII and V5.

Prerequisites:

• Python 3.5 and higher
• Keras framework with TensorFlow backend
• Numpy, Scipy, Pandas libs
• Scikit-learn framework

Instructions for running the program

1. Execute the training2017.zip and MIT-BH.zip files into folders training2017/ and MIT-BH/ respectively
2. If you want to use 2D Convolutional Neural Network for ECG classification then run the file CNN_ECG.py with the following commands:

• If you want to train your model on the 2017 PhysioNet/CinC Challenge dataset:

python ECG_CNN.py cinc

• If you want to train your model on the MIT-BH dataset:

python ECG_CNN.py mit

3. If you want to use 1D Convolutional Neural Network for ECG classification then run the file Conv1D_ECG.py with the following commands:

python Conv1D_ECG.py 0.9 55 25 10

where 0.9 is a fraction of training size for full dataset, 55 is a first filter width, 25 is second filter width, 10 is a third filter width.

Additional info

Citation

If you use my repo - then, please, cite my paper. This is a BibTex citation:

@article{pyakillya_kazachenko_mikhailovsky_2017,
    author = {Boris Pyakillya, Natasha Kazachenko, Nick Mikhailovsky},
    title = {Deep Learning for ECG Classification},
    journal = {Journal of Physics: Conference Series},
    year = {2017},
    volume = {913},
    pages = {1-5},
    DOI={10.1088/1742-6596/913/1/012004},
    url = {http://iopscience.iop.org/article/10.1088/1742-6596/913/1/012004/pdf}
}

For feature extraction and hearbeat rate calculation:

• https://github.com/PIA-Group/BioSPPy (Biosignal Processing in Python)