Neural correlates of learning in a linear discriminant analysis brain-computer interface paradigm

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interfaces (BCI) show potential for improved control with practice.
• The underlying neural mechanisms of BCI learning are not fully understood.
• Investigating neural correlates of learning is crucial for BCI development.

Purpose of the Study:

• To investigate how neural activity changes during motor BCI use.
• To determine if neural representations become more distinct with practice.
• To correlate neural changes with improvements in BCI task performance.

Main Methods:

• Two Macaque monkeys controlled a mobile robotic platform using a BCI with a linear discriminant analysis (LDA) decoder.
• Neuronal firing patterns were recorded using microelectrode arrays.
• Changes in neural signals within the LDA's linear discriminant (LD) space were analyzed over time.
• Direction selectivity was quantified using permutation feature importance (FI).

Main Results:

• Neural representations in the LD space diverged for different output classes within sessions.
• This divergence led to reduced misclassification errors and improved task accuracy.
• Higher accuracy correlated with channels showing strong directional preference (high FI) and varied population codes.

Conclusions:

• Monkeys demonstrated the ability to modulate neural activity for improved BCI control within sessions.
• Intra-sessional variations in neural representations are important for BCI learning.
• Distinct neural representations contribute to enhanced BCI performance and accuracy.