Researchers led by Dr. Cong from the Department of Computer Science at UCLA have developed an algorithm that enables real-time control in brain-machine interface applications.
A brain-machine interface (BMI) is a device that allows a machine to be controlled by neural signals recorded from the brain. To function properly, a BMI must accurately detect specific neural signal events, and then reliably trigger an appropriate machine command in response to those events. Some BMI devices operate by monitoring electroencephalographic (EEG) signals recorded from the brain, and applying filters to detect specific neural events in the EEG data. However, such filters commonly incur signal detection delays lasting from a few milliseconds up to several seconds (depending upon what kind of neural signal is being detected by the filter).
These long filter delays can be problematic for BMI applications that require short signal detection latencies. For example, BMIs for closed-loop neurostimulation are increasingly being used to treat neurological disorders. These BMI devices detect abnormal patterns of brain activity (such as epileptic seizures or Parkinsonian tremors), and then rapidly deliver feedback stimulation to the brain that corrects the abnormal activity patterns, and thereby relieves symptoms of the disease. In some cases, neurofeedback must be synchronized in real time with ongoing rhythms in the patient’s brain. To provide the most effective treatment, these kinds of closed-loop neurofeedback devices must deliver stimulation as quickly as possible after abnormal patterns of brain activity are detected. Such devices must also be highly energy efficient, since they are often surgically implanted into patients and must run for days or weeks on a single battery charge.
Researchers led by Dr. Cong from the Department of Computer Science and Hugh T. Blair in the Department of Psychology at UCLA have developed an algorithm that facilitates real-time control in brain-machine interface (BMI) applications. Their invention utilizes artificial neural networks that can be trained to predictively filter EEG recordings, and thereby detect specific neural signals almost instantaneously (within microseconds) after they occur, rather than after long delays that are incurred by more traditional filtering methods. Increased signal detection speed is critical for certain BMI applications—such as generation of neurofeedback by closed-loop stimulator devices—where performance is constrained by how rapidly the BMI can generate control signals in response to detected brain events. The new algorithm is computationally efficient enough to be implemented on microchips that consume very little power, and can be scaled up to process hundreds or thousands of neural signals on a single device, making it ideal for use in chronic surgical implants which require long battery life.
Brain machine interface, neural networks, LSTM, deep brain stimulation, Parkinson’s, epilepsy, medical devices, implant, brain, neural recordings, closed-loop, real-time control