CombSleepNet

Hyeong-Jin Kim, Minji Lee, Dong-Ok Won, and Seong-Whan Lee, "CombSleepNet: Combination of Spectral–Temporal CNN and Bi-LSTM with Dense Connectivity for Automatic Sleep Stage Scoring," submit to IEEE Journal of Biomedical and Health Informatics.

How to run

1. Download the Sleep-EDF database
   • Sleep-EDF database is available here.
     Note that, we used only 20 people of the Sleep Cassette data.
     i.e. from subject SC4001 to SC4192 in the sleep-cassette.

2. Data pre-processing

   • Change directory to ./CombSleepNet/pre-processing
   • Unzip eeglab.zip
   • Run preprocessing.m

3. Training and testing the CombSleepNet

   • Change directory to ./CombSleepNet
   • Refer to example.txt to train and test CombSleepNet.

Environment:

• Matlab R2019b
• Python3
• Pytorch v1.3.1
• numpy v1.17.4
• scipy v1.3.3
• scikit-learn v0.22

Result:

Hypnogram and posterior probability distribution with CombSleepNet for one subject of Sleep-EDF database

Contact:

Hyeong-Jin Kim

Department of Brain and Cognitive Engineering, Korea University
Email: kme0115@korea.ac.kr

License:

This project is released under the MIT License.