Ultra-low power brain implants find a significant signal in gray matter noise

By tuning into a subset of brain waves, researchers at the University of Michigan have dramatically reduced the power requirements of neural interfaces while improving their accuracy, a discovery that could lead to durable brain implants that can treat neurological diseases and allow prosthetics. controlled by the mind and machines.

The team, led by Cynthia Chestek, associate professor of biomedical engineering and faculty at the Institute of Robotics, estimated a 90% drop in energy consumption from neural interfaces using their approach.

“Today, interpreting brain signals into someone’s intent requires computers as tall as people and lots of electrical power, worth several car batteries,” said Samuel Nason, first study author and doctor. candidate in the Chestek Cortical Neural Prosthesis Laboratory. “Reducing the amount of electrical energy by an order of magnitude will eventually allow brain-machine interfaces in the home.”

Neurons, the cells in our brains that transmit information and action around the body, are noisy transmitters. Computers and electrodes used to collect data from neurons are listening to a radio trapped between stations. They must decipher the actual content amid the hum of the brain. Complicating this task, the brain is a hose for this data, increasing power and processing beyond the limits of safe implantable devices.

Currently, to predict complex behaviors, such as grasping an object for neural activity from a hand, scientists can use transcutaneous electrodes or direct wiring through the skin to the brain. This can be accomplished with 100 electrodes that capture 20,000 signals per second, and allows for feats like re-activating a paralyzed arm or allowing someone with a prosthetic hand to feel how hard or soft an object is. But this approach is not only impractical outside of the laboratory setting, it also carries a risk of infection.

Some wireless implants, created with highly efficient application-specific integrated circuits, can achieve nearly the same performance as transcutaneous systems. These chips can collect and transmit around 16,000 signals per second. However, they still have to achieve consistent operation, and their custom nature is an obstacle to obtaining approval as safe implants compared to industrially manufactured chips.

“This is a great leap forward,” said Chestek. “To get the high-bandwidth signals that we currently need for brain machine interfaces wirelessly, it would be completely impossible given the power supplies of existing pacemaker devices.”

To reduce energy and data needs, researchers compress brain signals. Focusing on spikes in neural activity that cross a certain power threshold, called the threshold crossing rate, or TCR, means that less data must be processed while predicting activated neurons at the same time. However, TCR requires listening to the entire pumping hose for neural activity to determine when a threshold is crossed, and the threshold itself can change not just from one brain to another but in the same brain on different days. This requires adjusting the threshold and additional hardware, battery, and time to do so.

By compressing the data in another way, Chestek’s lab marked a specific feature of the neuron data: the power of the spike band. SBP is an integrated set of frequencies from multiple neurons, between 300 and 1,000 Hz. By listening to only this frequency range and ignoring others, taking data from a straw instead of a hose, the team found a very accurate prediction of behavior with dramatically lower energy needs.

Compared to transcutaneous systems, the team found that the SBP technique is just as accurate while receiving a tenth of the signals, 2,000 vs. 20,000 signals per second. Compared to other methods, such as using a threshold crossing rate, the team’s approach not only requires much less raw data, but is also more accurate in predicting firing of neurons, even in noise, and does not require adjust a threshold.

The team’s SBP method solves another problem that limits the life of an implant. Over time, the electrodes on an interface cannot read the signals between the noise. However, because the technique works just as well when a signal is half what is required of other techniques such as threshold crossings, the implants may be left in place and used longer.

While new brain-machine interfaces can be developed to take advantage of the team method, their work also unlocks new capabilities for many existing devices by reducing the technical requirements to translate neurons into intent.

“It turns out that many devices have been sold short,” said Nason. “These existing circuits, which use the same bandwidth and the same power, are now applicable to the entire field of brain-machine interfaces.”

The study, “A low-power band of neural activity dominated by local individual units improves the performance of brain-machine interfaces,” is published in Biomedical Engineering of Nature.