/Augmentation-of-ECG-Training-Dataset-with-CGAN

Imbalanced MIT-BIH dataset is augmented with the generated beats by WGAN-GP and AC-WGAN-GP to improve the classification metrics.

Primary LanguagePython

paper: Arrhythmia Classification using CGAN-augmented ECG Signals

arXiv: https://arxiv.org/abs/2202.00569

please cite as:
@article{adib2022arrhythmia,
title={Arrhythmia Classification using CGAN-augmented ECG Signals},
author={Adib, Edmond and Afghah, Fatemeh and Prevost, John J},
journal={arXiv preprint arXiv:2202.00569},
year={2022}
}

the read/save paths in the file ("main_ac_wgan_gp_ecg.py") should adjusted according to your filing system

download the MITBIH dataset file ("record_X_y_adapt_win_bef075_aft075_Normalized.json") from Google Drive via the following link and place it in the proper folder accordingly:

https://drive.google.com/file/d/1d2gUuhJeWwVtKfzPbZx9gJgeBhXp3ibb/view?usp=sharing

raw signals (from PhysioNet website https://physionet.org/content/mitdb/1.0.0/) are segmented according to an adaptive window scheme: heart rate (R-R distance) is calculated at each beat and 75% of the distance before and after each R-peak are used as the cutoffs for segmentations. The individual beats are put in a dictionary. The keys are the record numbers and the values are lists of segmented beats.

link to all data in MITBIH Arhythmia dataset in 'dictionary' format: https://drive.google.com/file/d/1mVzEMRCzA-j3CFEgOvqGSQcQAbg0Gz4N/view?usp=drive_link

link to the 7 classes of real data used in this study: https://drive.google.com/file/d/1h0TtJAvJJK3IwDbqzK6pLKVBQw5VpvDL/view?usp=drive_link