Center of Excellence for Learning in Education, Science and Technology A National Science Foundation Science of Learning Center

Capstone Project 1: Adaptive Brain-Computer Interactions

Led by CELEST Co-PI member Frank Guenther

The unifying aim of ABCI, the Adaptive Brain-Computer Interactions CP, led by co-PI Frank Guenther, is to develop a reliable brain-computer interface (BCI) system that addresses the needs of severely paralyzed individuals. A typical BCI involves an electrophysiological recording system that routes neural signals (often reflecting motor commands) to a computer, where they are transformed into computer commands via a decoder that has been intelligently designed to predict user intent from the neural signals. The predicted intentions typically take the form of cursor movements or other computer commands. With practice, BCI users typically improve their performance due to motor learning and other forms of learning; that is, their brains adapt to produce more effective input signals to the BCI. Learning can also take place within the decoder by adapting decoder parameters in a way that reduces performance error. This decoder adaptation process can occur in parallel with human learning, thereby reducing the amount of practice needed to effectively use the BCI.

BCIs are ideally suited to restoring communication capabilities to severely paralyzed individuals such as those suffering from locked-in syndrome, characterized by total loss of voluntary movement with intact cognition and sensation. BCIs can also be used to control mobile robots (such as a robotic wheelchair) or robotic limbs. A number of research groups within CELEST and around the world have generated impressive demonstrations using BCI technologies in laboratory settings. However, to date there is no reliable BCI system available for home use by severely paralyzed individuals. CELEST is ideally poised to fill this void because it brings together a uniquely large, varied, and qualified team of computational neuroscientists, experimentalists, clinical scientists, and engineers whose efforts will be integrated and focused by this CP.

One of the biggest hurdles preventing the deployment of BCI technology in the real world can be summed up by the following central problem: although some people can learn to use current BCIs quite reliably, a significant proportion of people, particularly those with disabilities, fail to learn to use them effectively. The ABCI CP will attack this problem from multiple perspectives, thereby exemplifying CELEST’s primary purpose, which is to understand how the brain learns as a whole system and then to transfer this knowledge into new technologies. In summary, CELEST research within this CP is guided by the following research question:

How can we build brain-computer interfaces that are easy to learn to use and
that restore communication and other capabilities to paralyzed individuals?

In answering this question, we will synthesize knowledge from CELEST's four scientific initiatives. Knowledge regarding neural plasticity, particularly plasticity in the attention and motor systems, will guide choices regarding stimuli used in our BCI system as well as training protocols for learning to use the BCI in order to make the system easy to learn to use. knowledge of dynamic coding of information in the frontal eye fields and motor cortical areas (for intracranial electrodes) and visual cortical areas (for non- invasive BCIs using steady-state visually evoked potentials or SSVEPs and steady- state auditory evoked potentials or SSAEPs, read using EEG on the scalp) will allow us to produce better decoders for translating neural signals into computer commands, including decoders that take advantage of information encoded by the timing of neural events relative to large-scale neural rhythms. knowledge regarding processing bottlenecks within sensory processing, working memory, and motor control systems will help us identify stimuli that require relatively little effort to process; in other words, we will develop stimuli that minimize the load on brain areas where information processing load is already high during BCI use. Finally, knowledge of functional connections will allow us to build decoders that take into consideration long-range synchronies that have been shown to bear information regarding movement intent, attentional focus, and working memory content in prefrontal cortex.