In an important advance, the UC San Francisco Weil Institute for Neurosciences researchers working towards a brain-controlled prosthesis has shown that machine learning technology allows a person to use their brain activity without the need for extensive daily retention. All previous brain-computer interface (BCI) efforts have been required to help control the computer cursor while doing so.
“The BCI sector has made a lot of progress in recent years, but because existing systems have had to reset and recalculate each day, they have not been able to tap into the brain’s natural learning processes.
It seems like a person has to learn to ride a bike repeatedly from scratch, ”said senior author Karunesh Ganguly, MD, PhD, an associate professor in the UCSF Department of Neurology.
The achievement of the “plug and play” display demonstrates the value of so-called ECOG electrode arrays for BCI applications. An ECoG array consists of a pad of electrodes, about the size of a note that is surgically placed on the surface of the brain. They allow long-term, stable recording of neural activity and have been approved for seizure monitoring in patients with epilepsy.
In contrast, previous BCI efforts have used “pin-cushion” style arrays of sharp electrodes that penetrate the brain tissue for more sensitive recording but shift or lose signal over time.
In this case, the authors obtained investigational device approval for long-term chronic implantation of ECOG arrays in paralyzed subjects to test their safety and efficacy as long-term, stable BCI implants.
In their new paper, published on September 7, 2020 in Nature Biotechnology, Ganguly’s team documented the use of an ECOG electrode array in a person with paralysis of all four organs (tetraplegia).
The participant is also enrolled in a clinical trial designed to test the use of ECoG arrays to allow paralyzed patients to control the prosthetic arm and arm, but in the new paper, the participant Used implants to control a computer cursor on a screen.
The researchers developed a BCI algorithm that uses machine learning to match the brain activity recorded by the ECOG electrode to the user’s desired cursor movements. Initially, the researchers followed the standard practice of resetting the algorithm every day.
The participant will begin by imagining specific neck and wrist movements while moving the cursor over the screen. Gradually the computer algorithm will update itself to match the movements of the cursor which is for this generated user, the effective passing control user of the cursor.
However, starting this process every day can lead to a serious limitation on the level of control that can be achieved. The device may take hours to control, and in a few days the participant had to give up completely.
The researchers then switched to allow the algorithm to continue to update to match the participant’s brain activity without resetting the algorithm every day.
They found that the continuous interval between brain signals and machine learning-algorithm resulted in a continuous improvement in performance for several days. Initially there was little lost ground to make each day, but soon the participant was able to quickly achieve top level performance.
“We found that we could further improve learning by making sure that the algorithm was not updating faster than the brain – a rate of about once every 10 seconds,” said Ganguly, UCSF Health and San Francisco veterans With a neurologist administration medical center’s neurology and rehabilitation service.
“We see it as an attempt to create a partnership between two learning systems – the brain and the computer – that eventually lets the artificial interface become an extension of the user’s own hand or arm.”
Over time, the participant’s brain was able to increase patterns of neural activity that could most effectively use an artificial interface via the ECoG array, eliminating less effective signals – a complex process such as the brain’s own Complex tasks are thought to be learned. Researchers say.
They observed that the activity of the participant’s brain appears to develop a restrained and coherent mental “model” to control the BCI interface, something that had never happened with daily resetting and recalculation.
When the interface was reset after continuous learning for several weeks, the participant rapidly reestablished the same pattern of neural activity to control the device – effectively retracting the algorithm to the former position.
Finally, once the expertise is established, the researchers show that they can close the need to update the algorithm completely, and the participant can start using the interface only daily, again without any need. To use or recalculate.