A general dual-pathway network for EEG denoising.

Area of Science:

• Neuroscience
• Signal Processing
• Artificial Intelligence

Background:

• Scalp electroencephalogram (EEG) signals are vital for brain activity analysis but are often weak and corrupted by artifacts.
• Deep learning models show promise for EEG denoising but can be large-scale and prone to overfitting.
• Existing methods struggle with the complexity and variability of EEG noise.

Purpose of the Study:

• To introduce a novel dual-pathway autoencoder (DPAE) modeling framework for effective scalp EEG signal denoising.
• To demonstrate the superiority of the DPAE model compared to traditional deep learning architectures like MLP, CNN, and RNN.
• To validate the denoising performance of DPAE on benchmark EEG artifact datasets.

Main Methods:

• Development of a dual-pathway autoencoder (DPAE) architecture for EEG signal processing.
• Comparative analysis of DPAE against Multi-layer Perceptron (MLP), Convolutional Neural Network (CNN), and Recurrent Neural Network (RNN) models.
• Validation using established scalp EEG artifact datasets to assess denoising efficacy.

Main Results:

• The DPAE model significantly reduces computational requirements compared to existing deep learning algorithms.
• DPAE demonstrates superior denoising performance, outperforming other models in root relative mean square error (RRMSE) metrics.
• Effective noise reduction was achieved in both time and frequency domains of the EEG signals.

Conclusions:

• The DPAE architecture offers an efficient and high-performing solution for scalp EEG signal denoising.
• This general network model is suitable for blind source separation without prior noise distribution knowledge.
• DPAE represents a significant advancement in processing noisy EEG data for improved brain activity analysis.