Skip to main content
placeholder image

Classification of Sleep Arousal using Compact CNN

Conference Paper


Abstract


  • © 2020 IEEE. Sleep arousal is a common health problem that negatively affects the quality of sleep. This study investigates the use of a compact convolutional neural network (CNN) to classify apnea and non-apnea sleep arousal categories. The experiments are conducted on a randomly selected subset of the physiological signals provided by the PhysioNet 2018 challenge dataset. In particular, three electroencephalography (EEG) channels, two electromyography (EMG) channels, electrooculography (EOG), and airflow data are used to build the classification model. Physiological signals are down-sampled by a factor of 2 and then split into two-second long non-overlapping window segments. A data augmentation technique is then applied to overcome the large class imbalance ratio between two sleep arousal categories. The network is trained on 80% of the segments extracted from the data of 100 subjects. With only 594 trainable parameters, our approach achieves an area under the precision-recall curve (AUPRC) of 0.677 for the intra-subject test (20% of the data from the 100 subjects), and 0.183 on the inter-subject test on the data of another 12 unseen test subjects. This result falls within the range of the official scores of the challenge winners, indicating a promising application in using this lightweight CNN model for automated classification of sleep arousal.

Authors


  •   Eldaraa, Ahmed (external author)
  •   Baali, Hamza (external author)
  •   Bouzerdoum, Salim
  •   Belhaouari, Samir (external author)
  •   Alam, Tanvir (external author)
  •   Rahman, Anas (external author)

Publication Date


  • 2020

Citation


  • A. Eldaraa, H. Baali, A. Bouzerdoum, S. Belhaouari, T. Alam & A. Rahman, "Classification of Sleep Arousal using Compact CNN," in 2020 IEEE International Conference on Informatics, IoT, and Enabling Technologies, ICIoT 2020, 2020, pp. 247-253.

Scopus Eid


  • 2-s2.0-85085515847

Ro Metadata Url


  • http://ro.uow.edu.au/eispapers1/4048

Start Page


  • 247

End Page


  • 253

Abstract


  • © 2020 IEEE. Sleep arousal is a common health problem that negatively affects the quality of sleep. This study investigates the use of a compact convolutional neural network (CNN) to classify apnea and non-apnea sleep arousal categories. The experiments are conducted on a randomly selected subset of the physiological signals provided by the PhysioNet 2018 challenge dataset. In particular, three electroencephalography (EEG) channels, two electromyography (EMG) channels, electrooculography (EOG), and airflow data are used to build the classification model. Physiological signals are down-sampled by a factor of 2 and then split into two-second long non-overlapping window segments. A data augmentation technique is then applied to overcome the large class imbalance ratio between two sleep arousal categories. The network is trained on 80% of the segments extracted from the data of 100 subjects. With only 594 trainable parameters, our approach achieves an area under the precision-recall curve (AUPRC) of 0.677 for the intra-subject test (20% of the data from the 100 subjects), and 0.183 on the inter-subject test on the data of another 12 unseen test subjects. This result falls within the range of the official scores of the challenge winners, indicating a promising application in using this lightweight CNN model for automated classification of sleep arousal.

Authors


  •   Eldaraa, Ahmed (external author)
  •   Baali, Hamza (external author)
  •   Bouzerdoum, Salim
  •   Belhaouari, Samir (external author)
  •   Alam, Tanvir (external author)
  •   Rahman, Anas (external author)

Publication Date


  • 2020

Citation


  • A. Eldaraa, H. Baali, A. Bouzerdoum, S. Belhaouari, T. Alam & A. Rahman, "Classification of Sleep Arousal using Compact CNN," in 2020 IEEE International Conference on Informatics, IoT, and Enabling Technologies, ICIoT 2020, 2020, pp. 247-253.

Scopus Eid


  • 2-s2.0-85085515847

Ro Metadata Url


  • http://ro.uow.edu.au/eispapers1/4048

Start Page


  • 247

End Page


  • 253