Cascaded Thinning in Upscale and Downscale Representation for EEG Signal Processing

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Electroencephalogram (EEG) signal processing commonly employs smoothing filters for noise reduction.
• Existing methods often struggle to balance noise removal with the preservation of crucial signal features.
• Upscale and Downscale Representation (UDR) offers a novel approach by converting signals to image domains for processing.

Purpose of the Study:

• To propose and evaluate Cascaded Thinning Upscale and Downscale Representation (CTUDR) for enhanced EEG signal smoothing.
• To compare CTUDR's performance against conventional filtering techniques.
• To assess CTUDR's effectiveness in both signal smoothing and classification tasks.

Main Methods:

• EEG signals are visualized and converted into binary images using UDR.
• Morphological operations are employed in a cascaded thinning process within the image domain.
• CTUDR performs a two-stage upscale and downscale operation on a single binary image representation.
• The processed signal is projected back to the time domain.

Main Results:

• CTUDR demonstrated superior performance in signal smoothing, achieving the best fitting error and correlation coefficient.
• When applied as a pre-processing step for EEGNet, CTUDR significantly improved classification accuracy (0.7640%) and F-measure (0.7607%).
• CTUDR outperformed Moving Average, Binomial, Median, and Savitzky Golay filters in both smoothing and classification tasks.

Conclusions:

• CTUDR offers an effective and efficient method for smoothing EEG signals, outperforming traditional filters.
• The image-domain processing approach of CTUDR preserves essential signal features while reducing noise.
• CTUDR shows significant potential for improving the performance of EEG-based machine learning models.