Simultaneous multiple-stimulus auditory brain-computer interface with semi-supervised learning and prior probability...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Auditory brain-computer interfaces (BCIs) allow command selection via brain activity from auditory stimuli.
• Existing auditory BCI paradigms face limitations in increasing command numbers without reducing selection speed due to sequential stimulus presentation.
• The standard oddball paradigm requires independent, sequential stimulus presentation, hindering scalability.

Purpose of the Study:

• To develop a novel double-stimulus auditory BCI paradigm capable of presenting multiple auditory stimuli simultaneously.
• To enhance BCI performance by integrating semi-supervised learning (SSL) and prior probability distribution tuning.
• To evaluate the impact of the new paradigm and SSL on BCI accuracy and information transfer rate (ITR).

Main Methods:

• A double-stimulus paradigm was developed, presenting simultaneous sounds from left and right auditory fields.
• Six distinct sounds were utilized for a 6x6 letter matrix, enabling a larger command set.
• Semi-supervised learning (SSL) was employed to update classifier weights, with prior probability distributions tuned using uniform, empirical, and extended empirical (e-empirical) methods.

Main Results:

• The double-stimulus paradigm without SSL achieved 67.89% accuracy and 2.67 bits/min ITR.
• Integration of SSL with the e-empirical distribution significantly improved performance to 74.59% accuracy and 3.37 bits/min ITR.
• Event-related potential analysis indicated contralateral and right-hemispheric dominance contributed to performance gains.

Conclusions:

• The proposed double-stimulus auditory BCI paradigm effectively increases the number of selectable commands.
• Incorporating SSL and e-empirical prior distribution optimizes BCI performance, enhancing accuracy and typing speed.
• This approach offers a viable solution for expanding auditory BCI capabilities without compromising usability.