for those who have lost their hearing, and communication devices for those who are paralyzed. Today, researchers are demonstrating real-world restorative BCI systems, both invasive and noninvasive, that are giving paralyzed individuals more effective ways to interact with their environment and even move.

The complex issues that scientists deal with in this area are numerous, and include challenges ranging from practical lab logistics, sensor hardware, and data-processing systems to team members who have come to the work from very dissimilar disciplines, making it difficult for the field to establish and maintain consistent terminology. For Dawn Taylor, a research scientist working in this area, such problems are not insurmountable. “With a close-knit lab, everyone learns a common vocabulary fairly quickly,” says Taylor, who conducts her research at the Cleveland Clinic Department of Neurosciences and the Cleveland VA Functional Electrical Stimulation Center.

Despite the ongoing challenges, research in BCI technologies is resulting in real help for people with severe disabilities, says Taylor. In her current work, Taylor is focusing on using brain signals to trigger movement

Despite the ongoing
challenges, research
in Bci technologies
is resulting in real
help for people with
severe disabilities,
says Dawn taylor.

in paralyzed individuals, bypassing damage in the spinal cord. “We are getting pretty good at decoding someone’s intended arm and hand movements from recorded brain activity,” she says, noting that her colleagues are now able to restore arm and hand function in paralyzed individuals by using implanted stimulators to activate paralyzed muscles. The main goal for her at this point is to link these two technologies—on the one hand effectively decoding the brain’s movement intentions and on the other hand successfully stimulating paralyzed muscles to produce movement—to enable paralyzed people

neural activity from a 16-channel electrode array for a Bci system. each spike indicates a firing neuron. the software determines which neuron generates each detected spike by attributing each spike to the neuron whose wave shape is most similar to its own.

to move their limbs just by thinking about doing so.

To devise a brain-controlled typing system for a patient who is unable to communicate, one method would be to display a keyboard on a screen and have the patient think about moving his or her fingers to each letter. While it would be possible to decode the movement path the patient is imagining and use that data to control a mouse on the screen, Taylor says it might be more efficient to decode the final goal of each pointing movement and select each letter directly. Or, Taylor explains, decoding each muscle’s activation level might be the most efficient strategy when using the brain to control an implanted stimulation system that activates paralyzed muscles to restore arm motion.

Depending on where the electrodes are implanted in the brain’s informa-tion-processing stream, BCI researchers can decode the goal of the movement itself, the trajectory in 3D space that a hand follows during a movement, the angular motion of the joints, or even the force in each muscle. The type of device that the BCI system controls will impact which aspect of movement to decode and where it would be best to acquire data from the brain’s processing stream.

What is clear from Taylor’s research and other BCI studies is that users can improve their device-movement skills with practice. “The brain is an adaptable learning machine,” says Taylor. “We learn to do a new dance move or play tennis through practice. Learning to control the movements of a device directly with the brain is no different.” Taylor says such learning can be accelerated by using smart algorithms that learn in parallel with the brain. Taylor’s latest algorithm, which she calls “co-adaptive,” is designed to decode muscle activation levels from the brain to restore arm and hand function via implanted stimulators. The algorithm, which must be set up in an initial supervised training phase where it is clear what the patient is trying to accomplish, is designed to modify itself on the basis of how accurate recent past movements were. Taylor likens this update process to how supervised learning occurs in neural-network algorithms.