EEG-Infinity: A Mathematical Modeling-Inspired Architecture for Addressing Cross-Device Challenges in Motor Imagery.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalogram (EEG) data exhibits significant cross-device variability due to differences in brain-computer interface hardware.
• This variability poses challenges for EEG decoding and the standardized use of EEG datasets.
• Existing studies have not fully addressed the complexities of cross-device variability in EEG.

Purpose of the Study:

• To introduce a novel approach for modeling cross-device variability in EEG data.
• To propose a new deep domain adaptation network, EEG-Infinity, to overcome EEG decoding challenges.
• To lay the foundation for large-scale EEG model research by addressing data variability.

Main Methods:

• Developed a "sequentially comprehensive formula" and a "spatial comprehensive formula" to model cross-device variability.
• Proposed EEG-Infinity, a deep domain adaptation network featuring replaceable EEG feature extraction backbones and an "alignment head".
• Conducted systematic experiments across four EEG-based motor imagery datasets under 48 diverse cases.

Main Results:

• EEG-Infinity demonstrated superior performance compared to commonly used approaches.
• Achieved an average classification accuracy improvement of 1.51% across 34 experimental cases.
• The proposed method effectively mitigates cross-device variability in EEG data.

Conclusions:

• EEG-Infinity offers an effective solution for EEG decoding despite cross-device variability.
• The novel modeling approach and network architecture advance the field of EEG analysis.
• This work facilitates the development of more robust and standardized large-scale EEG models.