A time-frequency denoising method for single-channel event-related EEG

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Electroencephalogram (EEG) signals are susceptible to noise from sources like ECG, EOG, and EMG.
• Noise interference significantly hinders the accurate study and analysis of brain signals.
• Improved methods for EEG noise reduction are crucial for precise brain signal measurement.

Purpose of the Study:

• To develop and evaluate an advanced time-frequency denoising framework for EEG signals.
• To enhance the accuracy of brain signal analysis by effectively removing various noise interferences.
• To validate the proposed method against existing denoising techniques.

Main Methods:

• A novel time-frequency denoising framework combining short-time Fourier transform (STFT), bidimensional empirical mode decomposition (BEMD), and non-local means (NLM).
• Utilizing BEMD to decompose time-frequency representations into sub-signals for multi-scale noise removal.
• Applying NLM for structural self-similarity-based smoothing to preserve signal integrity.

Main Results:

• The proposed method effectively suppresses high-frequency noise, yielding smoother EEG waveforms.
• Correlation analysis showed the time-frequency denoised signal most closely resembled the reference signal compared to other methods (EEMD-ICA, EEMD-CCA, wavelet thresholding).
• This indicates superior noise reduction and signal fidelity.

Conclusions:

• The developed time-frequency denoising method demonstrates high efficacy in reducing noise in EEG signals.
• The approach offers a feasible strategy for more accurate brain signal determination.
• The method shows significant potential for practical applications in neuroscience and clinical settings.