The official implementation of the paper "Reconstructing QRS Complex from PPG by Transformed Attentional Neural Networks" https://ieeexplore.ieee.org/document/9109576

Graph Abstract

Model Architecture

Download the dataset:

https://drive.google.com/file/d/15dxbpi4FH7lJbRFZwyREyX4V0VKDxXNs/view?usp=sharing

And follow the instruction:

mkdir data
unzip dataset.zip -d data

After that, you should have following data structure:

data/
├── bidmc
│   ├── bidmc_csv
│   ├── bidmc-filtered
│   ├── bidmc-filtered-test
│   └── bidmc-filtered-train
└── uqvitalsigns
    ├── uqvitalsignsdata
    ├── uqvitalsignsdata-test
    └── uqvitalsignsdata-train

The main dataset we used in this paper can be found at

The University of Queensland Vital Signs Dataset

Download the model weights and usually we put it in ./weights.

https://drive.google.com/file/d/10aYWNkgaGCz1zU6--kN3yaW6L_9BzkhQ/view?usp=sharing

You can check it yourself in requirements.txt

• Ubuntu 18.04
• python 3.6
• pytorch 1.2 ...

# in your environment with pip
pip install -r requirements.txt

All the training parameters are included in config files.

# run UQVIT dataset with full model
python3 train.py --flagfile config/UQVIT.cfg

# run UQVIT dataset with LSTM baseline model
python3 train.py --flagfile config/UQVIT_LSTM.cfg

# run BIDMC dataset with full model
python3 train.py --flagfile config/BIDMC.cfg

Please see EXAMPLE.md.

Simple result:

tensorboard --logdir logs

If you use this code for your research, please cite our papers.

@ARTICLE{ppg2ecg,
  author={H. -Y. {Chiu} and H. -H. {Shuai} and P. C. . -P. {Chao}},
  journal={IEEE Sensors Journal}, 
  title={Reconstructing QRS Complex From PPG by Transformed Attentional Neural Networks}, 
  year={2020},
  volume={20},
  number={20},
  pages={12374-12383},}