New approach reduces neural interface power requirements while improving accuracy

Reviewed by Emily Henderson, B.Sc.July 27, 2020

Researchers at the University of Michigan tuned to a subset of brain waves to dramatically reduce the power requirements of neural interfaces and improve their accuracy. This finding could enable long-term brain implants that allow both treatment of neurological disorders and mind control. Prostheses and machines.

A team led by Cynthia Chestec, associate professor of biomedical engineering and core faculty at the Robotics Institute, estimated that their approach would reduce neural interface power consumption by 90%.

Currently, interpreting brain signals to someone’s intention requires a computer as high as a person and a lot of power. An energy reduction of an order of magnitude will ultimately enable the brain-machine interface at home. “

First author of the study, Ph.D. candidate for Samuel Nason Chestek’s Institute of Cortical Neuroprosthesis

Neurons are the noisy transmitters of cells in our brain that relay information and behavior around our body. Computers and electrodes used to collect neuron data are listening to radio stuck between stations. They have to decipher the actual content in the growl of the brain. Complicating this task is the brain’s fire hose of this data, which pushes power and processing beyond the limits of secure implantable devices.

Currently, scientists can use transcutaneous electrodes or wire them directly through the skin to the brain to predict complex behaviors such as grabbing a hand item from neuronal activity. This can be achieved with 100 electrodes capturing 20,000 signals per second, allowing you to re-enable a paralyzed arm or someone with a prosthetic hand to feel the hardness or softness of an object. However, this approach is not only practical in non-laboratory environments, it also carries the risk of infection.

Some wireless implants, made using highly efficient application specific integrated circuits, can perform almost as well as transdermal systems. These chips can collect and transmit about 16,000 signals per second. However, they have not yet achieved consistent behavior, and their custom-made properties are an obstacle to their approval as safe implants compared to industrial tips.

“This is a big leap,” said Cestek. “Wirelessly extracting the high bandwidth signals currently required for brain-machine interfaces is completely impossible given the power supply of existing pacemaker-style devices.”

To reduce the need for power and data, researchers compress brain signals. Focusing on spikes of neural activity above a certain power threshold, called the threshold crossing rate or TCR, predicts which neurons are firing and requires less data to be processed. However, TCR requires listening to a complete firehorse of neuronal activity to determine when a threshold is crossed, not only the threshold itself changing from one brain to another, but the same. It may change from day to day. This requires threshold adjustments and additional hardware, battery, and time adjustments for that.

By compressing the data in another way, Chestk’s lab dialed in to a specific feature of neuronal data, the spiking band power. SBP is an integrated set of frequencies from multiple neurons between 300 and 1,000 Hz. By listening only to frequencies in this range, ignoring other frequencies, and capturing data from straws rather than hoses, the team made very accurate forecasts and significantly reduced power demand.

Compared to transdermal systems, the team found that the SBP method was more accurate, capturing 1/10th the number of 10,000 signals per 2,000 signals per second. Compared to other methods such as using threshold crossing rates, the team’s approach not only has much less raw data, but is also more accurate in predicting neuron firing during noise, and There is no need to adjust the threshold value.

The team’s SBP method solves another issue that limits the useful life of implants. Over time, the interface electrodes will not be able to read the signal between the noise. However, if the signal is half of what is required by other techniques such as threshold crossing, this technique works the same, leaving the implant in place and allowing longer use.

While new brain-machine interfaces can be developed to take advantage of team methods, their work also unlocks new features on many existing devices by reducing the technical requirements to transform neurons into intent. To do.

“It turns out that many devices sell themselves short,” Nason said. “These existing circuits that use the same bandwidth and power can now be applied to the entire area of ​​brain-machine interface.”