Driving event-related potential-based speller by localized posterior activities: An offline study

Area of Science:

• Neuroscience
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Event-related potential (ERP)-based brain-computer interfaces (BCIs) typically use multi-sensor recordings to capture widespread cortical activity.
• This approach increases sensor count and computational load, hindering the development of mobile and wearable BCIs.
• Localizing brain activity could reduce sensor needs and computational complexity, facilitating BCI popularization.

Purpose of the Study:

• To investigate the localization of brain activity using novel visual graphic stimuli for an ERP-based speller.
• To compare the performance of graphic stimuli-based ERP spellers with traditional character-flashing paradigms.
• To evaluate the impact of sensor settings (full, normal, localized) on classification accuracy and information transfer rates.

Main Methods:

• Participants performed a spelling task using both graphic stimuli and traditional character-flashing paradigms.
• Novel visual graphic stimuli were employed to induce specific, localized brain responses.
• Classification accuracies and information transfer rates were compared across different sensor settings (FS, NS, LS) and paradigms.

Main Results:

• Graphic stimuli induced localized brain activity concentrated in the posterior brain region, unlike traditional stimuli.
• The graphic stimuli paradigm achieved significantly better performance under localized sensor settings compared to the traditional paradigm.
• Localized posterior brain activities were sufficient to drive an ERP speller with comparable or superior performance to the traditional paradigm.

Conclusions:

• Novel graphic stimuli can effectively localize brain activity for ERP-based spellers.
• Localized brain activity is sufficient for high-performance ERP spellers, reducing the need for extensive sensor arrays.
• This approach enhances the feasibility of mobile and wearable BCIs by decreasing computational burden and sensor requirements.