Symbiotic brain-machine drawing via visual brain-computer interfaces

Area of Science:

• Neuroscience
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Brain-computer interfaces (BCIs) are transitioning from research to clinical and assistive applications.
• There is a growing need for non-invasive BCIs with reduced hardware requirements.
• Existing non-invasive BCIs often face limitations in speed and accuracy.

Purpose of the Study:

• To develop a non-invasive BCI system for reconstructing imagined visual shapes.
• To enhance BCI performance through adaptive visual stimuli and AI integration.
• To explore symbiotic human-AI interaction for improved BCI capabilities.

Main Methods:

• Utilized steady-state visual evoked potentials (SSVEPs) analysis.
• Implemented iterative selection of adaptive visual probes at varying flicker-frequencies.
• Employed Gabor-inspired or machine-learned policies for probe placement.
• Leveraged single-channel EEG data and stable diffusion models for image reconstruction and enhancement.

Main Results:

• Successfully reconstructed simple imagined shapes within approximately two minutes.
• Achieved a more than 5x increase in BCI bit-rates through human-AI interaction.
• Transformed reconstructed mental images into realistic visual representations using stable diffusion models.

Conclusions:

• The developed non-invasive BCI offers a promising alternative to implantable technologies.
• Symbiotic human-AI interaction significantly boosts BCI performance.
• This work provides a foundation for advanced AI-augmented brain-computer interface development.