Automated Sleep Staging in Epilepsy Using Deep Learning on Standard Electroencephalogram and Wearable Data

Area of Science:

• Neurology
• Biomedical Engineering
• Sleep Medicine

Background:

• Automated sleep staging using wearable devices offers potential for improved epilepsy management.
• Current methods often require cumbersome clinical equipment like electroencephalogram (EEG).

Purpose of the Study:

• To evaluate a deep learning model's performance in automated sleep staging using wearable data compared to standard EEG.
• To explore sleep architecture differences in epilepsy patients with and without seizures.

Main Methods:

• A deep learning model was used for sleep staging on 223 night-sleep recordings from 50 epilepsy patients.
• Data was collected using both hospital-based EEG and a wearable device.
• Model performance was compared against clinical expert scoring using Bland-Altman analysis and mixed-effect models.

Main Results:

• The model achieved moderate accuracy (Cohen's kappa 0.59 for EEG, 0.43 for wearable) versus clinical experts.
• Wearable data underestimated most sleep macrostructure parameters, except for N2 sleep.
• Epilepsy patients with seizures slept longer and spent more time in N2 sleep compared to those without seizures.

Conclusions:

• Wearable EEG and accelerometry show potential for sleep monitoring in epilepsy patients.
• The automated analysis approach requires further refinement for clinical implementation.
• Sleep monitoring may reveal differences in sleep patterns related to seizure activity.