Sleep Spindle Detection Using RUSBoost and Synchrosqueezed Wavelet Transform

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Sleep spindles are crucial electroencephalographic (EEG) waveforms for sleep medicine.
• Manual detection of sleep spindles is time-consuming and burdensome, even for experts.
• Existing automated methods face challenges like threshold tuning and imbalanced data.

Purpose of the Study:

• To develop an automated sleep spindle detection method that overcomes limitations of conventional approaches.
• To combine wavelet synchrosqueezed transform (SST) for feature extraction and random under-sampling boosting (RUSBoost) for imbalanced data handling.
• To reduce the workload associated with polysomnography (PSG) scoring.

Main Methods:

• Proposed a novel method, SST-RUS, integrating SST for time-frequency analysis of spindle waveforms.
• Employed RUSBoost, a machine learning framework designed to manage imbalanced datasets.
• Validated the method using the Montreal archives of sleep studies cohort 1 (MASS-C1) dataset.

Main Results:

• The SST-RUS method demonstrated effective handling of imbalanced data in sleep spindle detection.
• Achieved an F-measure of 0.70, with 76.9% sensitivity and 61.2% positive predictive value.
• Eliminated the need for manual threshold tuning, unlike conventional template matching methods.

Conclusions:

• The proposed SST-RUS method offers an efficient and automated solution for sleep spindle detection.
• This approach can significantly alleviate the burden of manual polysomnography (PSG) scoring in sleep medicine.
• The integration of SST and RUSBoost provides a robust framework for analyzing imbalanced physiological signals.