semi-supervised-seizure-prediction

EEG power spectra parameterization and adaptive channel selection towards efficient semi-supervised seizure prediction

This is code for the paper [EEG power spectra parameterization and adaptive channel selection towards efficient semi-supervised seizure prediction] which in the minor revision

@article{lihanyi1761,
  title={EEG power spectra parameterization and adaptive channel selection towards efficient semi-supervised seizure prediction},
  author={Hanyi Lia, Jiahui Liao, Hongxiao Wang, Chang’an A. Zhan and Feng Yang},
  journal={computers in biology and medicine},
  year={2024}
}

Requirements

• tensorflow==2.6.2
• python==3.8.13
• numpy==1.19.5
• scipy==1.10.1
• six==1.15.0
• mn==1.4.0
• pywt==1.0.6
• stockwell==1.0.6

How to run the code

1. To obtain the channel selection results based on PAPC, please run the code below the file 'PAPC'

2. Set the paths in *.json files. Copy files in folder "copy-to-CHBMIT" to your CHBMIT dataset folder.

3. Prepare preprocessed data for WGAN-GP training. A large storage is required.

   python3 main.py --mode save_STFT --dataset DATASET

   • DATASET can be CHBMIT or siena-scalp-eeg-database-1.0.0.
   • CHBMIT(https://physionet.org/content/chbmit/1.0.0/) and siena(https://physionet.org/content/siena-scalp-eeg/1.0.0/) are both common dataset which can be download.

4. Train model.

   
   python3 main.py --mode wgan-gp --dataset DATASET

* For comparison, MODEL can be dcgan,wgan,wgan_gp .

5. Leave-one-seizure-out cross-validation.
```console
python3 main.py --mode cvgan --dataset DATASET