S5: Self-Supervised Learning Boosts Sleep Spindle Detection in Single-Channel EEG via Temporal Segmentation.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Sleep spindles are key EEG waveforms during N2 sleep, linked to cognitive functions.
• Manual sleep spindle identification is subjective, slow, and lacks consistency, hindering research.
• Automated detection methods are needed to advance sleep spindle research.

Purpose of the Study:

• To introduce S5, an automated method for accurate sleep spindle detection.
• To overcome limitations of manual visual inspection in sleep spindle analysis.

Main Methods:

• S5 utilizes a novel encoder-decoder architecture for time-series segmentation.
• A two-stage training approach (pre-training and fine-tuning) enhances detection precision.
• The method was evaluated on public datasets, including the MODA dataset.

Main Results:

• S5 achieved robust and competitive performance on multiple datasets.
• The automated method surpassed average human expert performance on the MODA dataset.
• Exploratory analysis on a large unlabeled dataset confirmed physiological validity.

Conclusions:

• S5 provides an accurate and efficient automated solution for sleep spindle detection.
• The developed method accelerates sleep research by simplifying spindle analysis.
• An accessible graphical toolbox enhances the usability of S5 for researchers.