Deform-CNN

Deformable convolutional networks for electrocardiogram diagnosis using pytorch. This is the source code of the article "An End-to-End 12-Leading Electrocardiogram Diagnosis System based on Deformable Convolutional Neural Network with Good Anti-noise Ability"

Getting started

Training environment

• NVIDIA driver version 418.67
• CUDA Version 10.1
• Python version 3.6.8
• Pytorch version 1.4.0+cu101

Install required package

From the root of this repository, run

pip install -r requirements.txt

Download the dataset

We use CPSC-2018 as dataset. You can dowload from here Reprocessing the data and save the data in .npy format. Then place the data into folder ./dataset/DataSet250HzRepeatFill and place the label into folder ./dataset. For more detail, please refer to load_dataset.py.

Notice:In our experiment, we removed data with total length more than 30 seconds, reducing the sampling frequency to 250 Hz and repeating filling. For more detail please refer to our article

Start training

From the root of this repository, run

python dcnv2_ecg_train.py

Customization

If you need to adjust some parameters, you can use argument. For example: python dcnv2_ecg_train.py --epochs 20 --optimizer Adam For more arguments, please refer to dcnv2_ecg_train.py.

Additional Notes

Please contact us by creating an issue if you would like to use this project for commercial purposes.

Cite this

@article{Qin_2021,
	doi = {10.1109/tim.2021.3073707},
	url = {https://doi.org/10.1109%2Ftim.2021.3073707},
	year = 2021,
	publisher = {Institute of Electrical and Electronics Engineers ({IEEE})},
	volume = {70},
	pages = {1--13},
	author = {Lang Qin and Yuntao Xie and Xinwen Liu and Xiangchi Yuan and Huan Wang},
	title = {An End-to-End 12-Leading Electrocardiogram Diagnosis System Based on Deformable Convolutional Neural Network With Good Antinoise Ability},
	journal = {{IEEE} Transactions on Instrumentation and Measurement}
} 

Reference: https://github.com/4uiiurz1/pytorch-deform-conv-v2 (MIT License)