LRR-UNet: A Deep Unfolding Network With Low-Rank Recovery for EEG Signal Denoising.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalogram (EEG) signals are vital for brain-computer interfaces but prone to noise.
• Deep learning models offer high performance but lack interpretability.
• Traditional Low-Rank Recovery (LRR) methods provide interpretability but can be computationally intensive.

Purpose of the Study:

• To develop an interpretable deep learning model for EEG denoising.
• To combine the performance of deep learning with the interpretability of LRR.
• To enhance the utility of EEG signals in brain-computer interface research.

Main Methods:

• Proposed LRR-Unet, a deep unfolding network.
• Transformed traditional iterative LRR into a neural network architecture.
• Replaced computationally expensive Singular Value Decomposition (SVD) and sparse optimization with learnable neural network modules.

Main Results:

• LRR-Unet outperformed state-of-the-art models in removing ocular and electromyographic artifacts.
• Achieved superior quantitative and qualitative denoising performance.
• Improved results in downstream EEG classification tasks after preprocessing.

Conclusions:

• LRR-Unet offers an effective and interpretable solution for EEG denoising.
• Demonstrated superiority in denoising performance.
• Validated practical utility in enhancing downstream applications.