Epileptic Seizure Forecasting with Generative Adversarial Networks

Code to reproduce results reported in our paper published as:

Truong, N. D., L. Kuhlmann, M. R. Bonyadi, D. Querlioz, L. Zhou, and O. Kavehei (2019). "Epileptic Seizure Forecasting with Generative Adversarial Networks." IEEE Access, accepted with minor revisions.

Receiver operating characteristics (ROC) curves of seizure forecasting performance testing for different patients of the three datasets: (a) - the CHB-MIT sEEG dataset, (b) - the Freiburg Hospital iEEG dataset, and (c) - the EPILEPSIAE sEEG dataset. Each line corresponds to one patient. Above the green dash line: good performance; above the blue dash line: very good performance.

Requirements

• hickle==3.4.3
• six==1.12.0
• tensorflow_gpu==1.12.0
• scipy==1.0.1
• np_utils==0.5.10.0
• pandas==0.24.2
• stft==0.5.2
• numpy==1.11.0
• mne==0.11.0
• scikit_learn==0.21.3

How to run the code

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

2. Prepare preprocessed data for DCGAN training. A large storage is required.

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

• DATASET can be FB, CHBMIT or EpilepsiaSurf (auxiliary files for EPILEPSIA Surface dataset are not uploaded here, available upon request) .

3. Train DCGAN model.

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

4. Leave-one-seizure-out cross-validation.

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