An accessible and versatile deep learning-based sleep stage classifier.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Manual sleep scoring is time-consuming and subjective, impacting research and clinical diagnosis.
• Advances in machine learning and accessible data have spurred development of automatic sleep stage classifiers.
• Existing automatic classifiers often present significant usability challenges.

Purpose of the Study:

• To introduce the Greifswald Sleep Stage Classifier (GSSC), an accessible and high-performance automatic sleep staging tool.
• To overcome barriers to use associated with current state-of-the-art classifiers.
• To provide a reliable, open-source solution for automatic sleep staging in research and clinical applications.

Main Methods:

• Development of the Greifswald Sleep Stage Classifier (GSSC) as a free, open-source software.
• Training the GSSC on large, publicly available hand-scored polysomnographic datasets.
• Ensuring GSSC's adaptability to diverse electrode configurations and integration capabilities with brain-computer interfaces.

Main Results:

• The GSSC demonstrates ease of installation and use on standard computer hardware.
• High-performance sleep staging is achievable with the GSSC across various electrode setups, including portable systems.
• The GSSC achieves accuracy comparable to or exceeding current state-of-the-art classifiers and human experts.

Conclusions:

• The GSSC offers a practical, accurate, and accessible solution for automatic sleep staging.
• Its open-source nature and adaptability make it a valuable tool for researchers and clinicians.
• The GSSC facilitates reliable sleep analysis, supporting advancements in sleep disorder diagnosis and research.