Noninvasive neuroimaging and spatial filter transform enable ultra low delay motor imagery EEG decoding.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interface (BCI) systems translate neural signals into machine commands, offering advanced human-computer interaction.
• Scalp electroencephalogram (EEG) suffers from limited spatial resolution due to volume conduction, hindering precise signal localization.
• Low-latency decoding is critical for real-time BCI applications, especially in motor imagery tasks.

Purpose of the Study:

• To enhance the spatial resolution of EEG signals for noninvasive intracranial activity exploration.
• To develop a low-delay decoding framework for real-time BCI systems.
• To improve the accuracy and interpretability of motor imagery classification using EEG.

Main Methods:

• EEG conduction was modeled using anatomical templates to derive cortical EEG via dynamic parameter statistical mapping.
• Filter bank common spatial pattern was employed to generate spatial filter kernels, reducing feature extraction computation to a linear level.
• A neural network with band-spatial-time domain self-attention mechanisms was utilized for feature classification and selection.

Main Results:

• The proposed method achieved high accuracy in classifying four types of motor imagery EEG tasks.
• The decoding framework demonstrated notably low latency.
• The results indicated high physiological interpretability of the decoded features.

Conclusions:

• The developed decoding framework significantly enhances BCI performance for motor imagery.
• This approach facilitates the development of real-time, low-latency human-computer interaction systems.
• The method offers a promising solution for overcoming the spatial resolution limitations of scalp EEG.