Quantitative analysis of task selection for brain-computer interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Brain-computer interfaces (BCIs) offer a communication pathway for individuals with severe motor impairments.
• BCI performance is highly dependent on the specific tasks users engage with.
• Optimizing task selection is crucial for enhancing BCI usability and effectiveness.

Purpose of the Study:

• To quantitatively evaluate the impact of task selection on brain-computer interface (BCI) performance.
• To analyze the benefits of subject-specific task selection across multiple datasets and users.
• To investigate the transferability of optimal task-pair information across different days.

Main Methods:

• Analysis of task-pairs from multi-class BCI imagery movement tasks across three datasets.
• Large-scale evaluation involving 109 users to assess task selection benefits.
• Assessment of task-pair information transferability across days for individual subjects.

Main Results:

• Subject-dependent optimal task-pair selection can increase the number of users controlling a binary BCI by approximately 20% compared to a fixed task-pair.
• Optimal task-pairs identified on one day generally perform well on subsequent days.
• User learning positively influences optimal task-pair generalization, though inexperienced users require special consideration.

Conclusions:

• Task selection is a critical step for developing usable BCIs, supported by significant quantitative evidence.
• Findings advocate for adaptive methods in task selection for practical, online BCI applications.
• Further research is encouraged to explore adaptive task selection with larger sets of mental tasks.