By Amy Blumenthal
Brain-machine interface systems can restore movement to disabled patients by turning their thoughts into movement, for example allowing them to move computer cursors or operate robotic limbs. These systems implant an electrode array in the brain to collect the neural activity and use a mathematical algorithm to estimate the subject’s intended movement from this activity in real time.
“To make these systems clinically-viable, their performance—how fast and accurately they can be operated by subjects—needs to significantly improve,” says Maryam Shanechi, Viterbi Early Career Chair in the Ming Hsieh Department of Electrical and Computer Engineering, who works at the intersection of neuroscience and electrical engineering to develop neurotechnologies for treatment of neurological disorders.
However, the performance of neuroprosthetic algorithms, has plateaued since 2010, says Shanechi. “This creates the need to investigate and develop new designs to break through the existing plateau,” she says.
A new study by USC Professor Maryam Shanechi and UC Berkeley Professor Jose Carmena, and colleagues offers a new mathematical algorithm that significantly improves the speed and accuracy by which subjects can move a neuroprosthetic compared to existing systems. The study, “Rapid Control and Feedback Rates Enhance Neuroprosthetic Control,” appearing this month in Nature Communications, puts forth a new state-of-the-art algorithm for the neuroprosthetic field and helps pave the way for clinical viability.
This new algorithm was inspired by ideas from control theory, a branch of electrical engineering that can be used to investigate how movements are optimally controlled. The researchers set out to study whether allowing the brain to send control commands to the prosthetic at rapid rates, i.e., every millisecond, could improve neuroprosthetic performance. In order to test their hypotheses, they developed a new algorithm that for the first time, enabled subjects to operate the neuroprosthetic at rapid millisecond-by-millisecond time-scales.
The researchers recorded the subjects’ brain activity consisting of the spikes of 15-33 neurons. The spikes from each neuron indicate the time-points at which that neuron fires an action potential. They then used concepts from control theory to train and build a point process filter algorithm that modeled the spikes as a time-series of zeros and ones. A ‘zero’ at a given time indicated the absence of an action potential at that time. Similarly, a ‘one’ at a given time indicated the presence of an action potential at that time. By modeling the spikes in this way, their new algorithm allowed the subjects to move the position of a cursor on a computer display every millisecond unlike prior designs based on Kalman filter algorithms that moved at much slower time-scales, i.e., only once every 50-100 milliseconds.
Using this new neuroprosthetic, the researchers tested whether the increased control rates could improve subjects’ speed and accuracy in performing various cursor movement tasks on a computer screen, e.g., moving from one target location to another or avoiding obstacles during movements. The hope was that their novel approach would account for the speed at which the brain could put out commands to move the cursor.
What they found was that increasing the control rate by allowing the brain to rapidly send control commands to the cursor enhanced neuroprosthetic performance—enabling subjects to perform the task faster and more accurately. Moreover, increasing the feedback rate by rapidly displaying the cursor movement to the subject further improved their performance. These results showed that the subjects’ brains did in fact have the ability to exploit a faster rate of control and a faster rate of feedback than existing neuroprosthetics were allowing. Comparing to the standard Kalman filter algorithm, they showed that their new algorithm increased performance by 32 percent on average across months of experiments.
This new neuroprosthetic presents a major departure from previous designs.
“Operating a computer interface faster will provide paralyzed or “locked-in” patients with an effective way to communicate with the outside world, for example to surf the web or to type,” Shanechi says.
“This could significantly improve the quality of life and autonomy for millions of disabled patients.” she adds..
The research could also have implications for the control of higher-dimensional robotic arms. The next step forward is to move the research into human clinical trials.
“We are very excited about this research since it not only shows the possibility to improve neuroprosthetic performance over existing techniques, but also dissociates the reasons behind this performance improvement thus providing principled design guidelines for future neurotechnologies., says Shanechi.
This work was funded by National Science Foundation grant EFRI-M3C 1137267 (J.M.C.), Defense Advanced Research Projects Agency contract N66001-10-C-2008 (J.M.C.) and National Science Foundation CAREER Award CCF-1453868 (M.M.S.).
For this study no testing of animals was conducted at USC. USC is a leader in the ethical and humane use of animals for research and teaching and is fully accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care, International (AAALAC) and has an animal welfare assurance on file with the NIH Office of Laboratory Animal Welfare.
Published on January 9th, 2017
Last updated on January 10th, 2019