Reading the Surface of the Brain
A startup aims to develop a minimally invasive neural prosthesis for disabled patients.
A technology currently used to monitor epilepsy is being adapted into a neural interface for people who are paralyzed or have motor impairments from neurodegenerative disease. Neurolutions, a startup based in St. Louis, is developing a small, implanted device that translates signals recorded from the surface of the brain into computer commands.
The device is based on electrocorticography (ECoG), in which a grid of electrodes is surgically placed directly on the surface of the brain to monitor electrical activity. This technology is currently used for surgical planning in patients with uncontrolled epilepsy in order to find the origin of their seizures. But Eric Leuthardt and Dan Moran at Washington University School of Medicine, in St. Louis, and Gerwin Schalk at the Wadsworth Center, in Albany, NY, are developing a much smaller version that would be implanted long term to allow paralyzed patients to control a computer and perhaps prosthetic limbs and other devices.
“We’re extremely excited about these signals because they are really opening a whole new avenue for extracting information from the brain in humans,” says Schalk. “The nice thing about ECoG is that it targets a space that no other sensor technology has been in before.”
Most efforts to build neural interfaces have focused on either electroencephalography (EEG), a noninvasive technology that records electrical activity from the scalp, or electrodes implanted into the brain. ECoG represents an intermediate between the two: because it records directly from the brain, it can provide a higher level of control than EEG, which is susceptible to distortion as the signal travels through the skull and as the patient moves. In addition, ECoG’s position on the surface of the brain may present fewer health issues than electrodes that penetrate brain tissue.
Because ECoG is used in epilepsy patients, researchers have already been able to conduct proof of principle experiments on a much wider scale than has been done using other invasive technologies. Tests of more than 20 patients have shown that people can quickly learn to move a cursor on a computer screen using their brain activity. Researchers first ask patients to imagine performing a certain action, such as moving a computer cursor to the left. They then identify changes in the frequency of electrical signals that correlate with that movement and use those to control the computer. The patient learns to more precisely control his or her brain activity and hence more reliably performs the task within half an hour.
“With minimal learning efforts, we have been able to tune and train the system to recognize simple commands, like ‘up,’ and ‘down,’ and ‘left’ and ‘right,’” says Shawn Lunney, Neurolutions’ chief executive officer. Lunney estimates that patients can control a computer cursor with approximately 80 percent accuracy.
“With the results from our studies, it made sense to develop the company and the intellectual property in parallel with a true next-generational implant,” says Leuthardt, a neurosurgeon and former TR100 winner who is continuing his studies in patients. He hopes to show that they can achieve three-dimensional control, which would be required for the most basic control of a prosthetic arm.
The researchers have already developed a preliminary prototype that is much smaller than the ECoG device used in epilepsy patients. While seizure monitoring requires electrodes to cover a large part of the brain’s surface, a neural interface can run on signals recorded from just a small brain area in the brain. While current ECoG surgeries require the removal of a large chunk of skull, a more compact technology could be delivered through a small burr hole in the skull.
Moran is testing the smaller device, which is about the size of three or four stacked quarters, in monkeys to determine its long-term potential, as well as the optimal parameters for an ECoG-based neural interface. For example, the researchers need to determine the number of electrodes that can be packed into the smallest space and still record enough independent information from the brain to control a prosthetic arm with multiple degrees of freedom. (The shoulder has three degrees of freedom, and the elbow two, all of which must be controlled by independent signals in the brain.)
One of the biggest issues in developing neural prostheses is creating a device that can record signals from the brain reliably over time. Electrodes implanted into the brain typically function well for only six to twelve months because the immune system attacks the electrode, encapsulating it in tissue and degrading the quality of the recorded signal. (There have been instances of implanted electrodes lasting for years, however.)
“What we really want is for these devices to work for a decade, but at least three to five years,” says Moran. So far, the small ECoG devices have been implanted in monkeys for about eight months, so it’s not yet possible to compare long-term durability and safety to that of more deeply implanted electrodes. Moran believes that because the device does not penetrate brain tissue, it will be less susceptible to immune attack.
“At the end of the day, the question is whether they can retain a high quality of recordings for a long period of time,” says Joseph Pancrazio, a program director at the National Institute for Neurological Disorders and Stroke, a government funding agency, in Bethesda, MD, who was not involved in the research. “The jury is still out.” Pancrazio also questions whether the procedure is truly less invasive than that used to implant electrodes into the brain, a technique that is employed clinically for deep brain stimulation, a treatment for Parkinson’s disease.
Neurolutions’ next step will be to build a prototype for clinical testing. Lunney estimates that the company will have a working prototype within the next year to 15 months. In the envisioned prototype, the implanted electrodes would wirelessly send signals to an external device, perhaps in the form of a hat or headband, which would both power the internal component and interpret the incoming information. “We’ve done the wireless communications,” says Lunney. “The challenge is turning it into one little electronics package that can fit into a nineteen-by-seven-millimeter package.”
AI is here.
Own what happens next at EmTech Digital 2019.