SpindleX: A Multi-Scale Event Detection Window and CBAM Attention-Based Model for Sleep Spindle Detection.

Area of Science:

• Neuroscience
• Sleep Medicine
• Biomedical Engineering

Background:

• Sleep spindles are key EEG waveforms in NREM sleep stage 2, associated with various diseases.
• Existing detection methods face challenges in utilizing spindle properties, avoiding thresholds, and adapting object detection principles.

Purpose of the Study:

• To introduce SpindleX, a novel end-to-end algorithm for automated sleep spindle detection.
• To address limitations of current methods by effectively using spindle characteristics and avoiding manual threshold settings.

Main Methods:

• SpindleX employs four modules: local feature extraction, sequence feature extraction, an attention module, and localization/classification.
• It utilizes convolutional neural networks and bidirectional LSTMs for feature extraction, with an attention mechanism for enhanced signal analysis.
• Multi-scale event detection windows are used for precise spindle localization, accommodating varying durations.

Main Results:

• SpindleX demonstrated significant efficiency on the MASS sleep dataset.
• The algorithm achieved an average F1-score of 78.9% on public datasets, indicating high reliability.

Conclusions:

• SpindleX offers a reliable automated method for sleep spindle detection, outperforming existing approaches.
• This tool can reduce manual scoring time and subjectivity, supporting clinical applications like sleep quality assessment and disease prognosis analysis.