EEGDiR: Electroencephalogram denoising network for temporal information storage and global modeling through Retentive...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Electroencephalogram (EEG) signals are crucial for brain research and clinical diagnostics.
• Noise contamination in EEG signals hinders accurate analysis of brain activity.
• Traditional denoising methods are often less effective than advanced deep learning techniques.

Purpose of the Study:

• To adapt the Retentive Network architecture, originally for large language models (LLMs), for effective EEG signal denoising.
• To improve the precision of brain activity analysis and neurological disorder diagnosis through enhanced EEG signal quality.
• To introduce a novel signal embedding technique for processing unidimensional EEG data within the Retentive Network.

Main Methods:

• Deployment of the Retentive Network architecture for EEG denoising.
• Development of a signal embedding strategy to adapt EEG signals for the Retentive Network.
• Utilizing the Retentive Network's temporal structure alignment for time-series EEG data.
• Creation of a standardized, preprocessed dataset to facilitate deep learning research in EEG denoising.

Main Results:

• Successful application of the Retentive Network for EEG signal denoising.
• Demonstrated improvement in analyzing brain activity due to reduced noise.
• Enhanced accuracy in the diagnosis of neurological conditions.
• Provided a valuable, ready-to-use dataset for future deep learning studies in this field.

Conclusions:

• The Retentive Network, with a novel embedding technique, offers a powerful new method for EEG denoising.
• This approach significantly improves the quality of EEG data, benefiting neuroscience and clinical applications.
• The standardized dataset accelerates progress in deep learning for EEG analysis and neurological disorder diagnosis.