ReHA-Net: a ReVIN-hybrid attention network with multiscale convolution for robust EEG artifact removal in...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Electroencephalography (EEG) is crucial for brain activity monitoring but suffers from signal degradation due to artifacts.
• Ocular movements, muscle activity, and environmental noise commonly contaminate EEG signals, hindering accurate analysis.
• Existing denoising methods often struggle to balance artifact removal with the preservation of essential neural information.

Purpose of the Study:

• To develop a robust deep learning framework, ReHA-Net, for effective electroencephalography (EEG) signal denoising.
• To enhance the signal-to-noise ratio and quality of EEG data for improved downstream applications.
• To create a generalized denoising solution that maintains subject-specific neural features.

Main Methods:

• Implemented a U-Net-based encoder-decoder architecture incorporating three novel modules: Hybrid Attention, Multiscale Separable Convolution (MSC), and Reversible Instance Normalization (ReVIN).
• Hybrid Attention module integrates temporal, spatial, and frequency attention mechanisms to differentiate neural patterns from structured noise.
• MSC block utilizes dilated and parallel depth-wise separable convolutions for capturing diverse temporal dependencies, while ReVIN aids in cross-subject generalization.

Main Results:

• ReHA-Net demonstrated superior denoising performance, achieving a Peak Signal-to-Noise Ratio (PSNR) of 27.10 dB and an Signal-to-Noise Ratio (SNR) of approximately 17.06 dB.
• The framework achieved a high correlation coefficient of 0.976 with clean signals and a low Relative Root Mean Square Error (RRMSE) of 0.165.
• The model successfully reduced artifacts while preserving crucial neural activity, outperforming existing methods.

Conclusions:

• ReHA-Net provides a robust and effective solution for electroencephalography (EEG) artifact removal.
• The framework's ability to preserve neural signals makes it a valuable preprocessing tool for clinical and research applications.
• ReHA-Net shows significant potential for enhancing tasks such as seizure detection, cognitive state monitoring, and brain-computer interfaces.