A hybrid network based on multi-scale convolutional neural network and bidirectional gated recurrent unit for EEG...

Area of Science:

• Neuroscience
• Signal Processing
• Artificial Intelligence

Background:

• Electroencephalogram (EEG) signals contain vital brain information but are often corrupted by artifacts.
• Artifacts like electromyographic and electrooculographic interference compromise EEG data reliability.
• Current deep learning methods, particularly CNNs, struggle with multi-scale and time-dependent features for effective artifact removal.

Purpose of the Study:

• To introduce a novel hybrid deep learning network, MSCGRU, for enhanced EEG artifact denoising.
• To address the limitations of existing methods in capturing multi-scale and temporal EEG features.
• To improve the accuracy and reliability of EEG signal processing for neuroscientific research and clinical applications.

Main Methods:

• Developed a Multi-Scale Convolutional Gated Recurrent Unit (MSCGRU) network, an enhanced generative adversarial network.
• Incorporated a multi-scale convolution module for extracting diverse frequency features from EEG.
• Utilized a channel attention mechanism to refine feature discriminability and a Bidirectional Gated Recurrent Unit (BiGRU) for temporal dependency extraction.

Main Results:

• MSCGRU demonstrated superior performance in denoising electromyographic artifacts compared to other models.
• Achieved a relative root mean square error of 0.277±0.009 and a correlation coefficient of 0.943±0.004.
• Obtained a signal-to-noise ratio of 12.857±0.294, indicating significant improvement in signal quality.

Conclusions:

• The proposed MSCGRU network effectively reconstructs clean EEG signals by addressing limitations in multi-scale and temporal feature extraction.
• MSCGRU outperforms existing artifact removal models, offering a promising advancement in EEG signal processing.
• This method has the potential to significantly benefit EEG-based diagnosis and treatment strategies.