An EEG signal smoothing algorithm using upscale and downscale representation

Area of Science:

• Signal Processing
• Neuroscience
• Biomedical Engineering

Background:

• Effective electroencephalogram (EEG) signal smoothing is crucial for accurate analysis and brain-computer interface (BCI) applications.
• Maintaining original signal features during smoothing is a significant challenge in EEG analysis.
• Cognitive conflict (CC) processing using EEG requires robust signal pre-processing techniques.

Purpose of the Study:

• To propose a novel EEG signal-smoothing algorithm that preserves signal features.
• To evaluate the algorithm's effectiveness in processing cognitive conflict (CC) tasks.
• To assess the algorithm's impact on classification accuracy and robustness in EEG analysis.

Main Methods:

• The proposed algorithm visualizes EEG signals by increasing line width, converting the representation frame into a binary image.
• An effective thinning algorithm is employed to obtain a unit-width skeleton, serving as the smoothed signal.
• The algorithm's application in CC processing is evaluated using classification and visual inspection tasks.

Main Results:

• The algorithm demonstrates high effectiveness in data fitting, particularly at high noise levels (SNR ≤ 5 dB), with fitting errors of 86.4%-90.4% compared to counterparts.
• Pre-processing EEG data with this algorithm significantly boosted the F1 score of state-of-the-art models by over 1%.
• Visual inspection tasks confirmed the algorithm's robustness, enabling clear observation of CC peaks like prediction error negativity and error-related positive potential (Pe).

Conclusions:

• The proposed EEG signal-smoothing algorithm offers significant advancements in signal processing for BCI applications.
• The algorithm enhances classification accuracy and provides robust identification of cognitive conflict-related neural signals.
• This approach represents a valuable tool for improving EEG analysis, particularly in noisy environments and for specific cognitive event detection.