SleepSpectralAE: Time-Frequency Autoencoder for Automated Sleep Disorder Screening

Authors

  • Swati Patil Department of Computer Applications, GH Raisoni College of Engineering and Management, Jalgaon, India
  • Sonal Patil Department of Computer Applications, GH Raisoni College of Engineering and Management, Jalgaon, India

Keywords:

— time-frequency ,sleep segments, health Disorders.

Abstract

An AutoEncoder for Time-Frequency Analysis of EEG Data to Help Automate the Screening Process for Sleep Disorders. SleepSpectralAE, a lightweight autoencoder based on the time-frequency domain of EEG was developed for the purpose of automating sleep disorder screening. The autoencoder creates an encoding of multichannel spectrograms into a low dimension, or latent space, that captures noise as well as subtle differences in the EEG signal related to Sleep Physiology abnormal phenomenon. The AutoEncoder was trained and evaluated using records of overnight polysomnography consisting of different types of sleep pathology including obstructive sleep apnea and periodic limb movement disorder. Results of the numerical experiments demonstrate that the reconstruction errors in the latent space of SleepSpectralAE provide a means of differentiating normal sleep segments from those associated with Sleep Pathology. The AUC for event detection of apnea events was 0.91 and a decrease in the false positive rate of 27% was observed compared to the baseline convolutional AE. These findings suggest that time-frequency autoencoding is a unique way to initially screen for Sleep Disorders in either a clinical or home-monitoring environment.

References

.

Published

2026-09-29