Translating neural signals into movement
by David Aum
A neural system so sophisticated that it can translate thoughts into prosthetic movements is being researched by Professor Krishna Shenoy of StanfordÕs Department of Electrical Engineering. ShenoyÕs Neural-Prosthetic Systems Laboratory (NPSL) works on translating brain signals accurately into commands that control prosthetics. Ultimately, their goal is to build a brain-computer interface for devices that patients with paralysis or amputation can control with their minds alone.
Gathering signals from neural implants
The first step in designing a thought-controlled prosthetic is to collect the brainÕs electrical impulses as voltage signals. Shenoy focuses on electrodes placed on top of the cerebral cortex beneath the skull. ÒThe systems we work on actually surgically implant electrodes on the brain," he says. "The alternate version is an electroencephalogram (EEG), which are basically scalp electrodes.Ó The main difference between the two is the strength and clarity of the signal. Shenoy describes the contrast with the following metaphor: ÒIt would be like listening to a microphone in a stadium. You would hear the crowd but if you could sit that microphone right up against people, you could hear the actual conversation. ThatÕs what we do by implanting the electrodes in the brain.Ó Shenoy notes that the bulky, wired implants of today must be wireless, unobtrusive, and built to last a lifetime to be truly feasible.
Depending on the purpose of the prosthesis, the areas of the cortex responsible for functions ranging from motion to touch are targeted. Signals from a specific region are collected, amplified, filtered to reduce stray electrical noise, and then converted into a digital format.
Understanding what the neural signals mean
Next, the digitized input signals must be analyzed. The electrical activity of each neuronÕs action potential conveys a message that must be combined across hundreds of neurons and translated in real time. How does one translate what a neuron is saying? Shenoy offers another helpful analogy: ÒLetÕs say that weÕre watching a football game with microphones in front of us. As the game goes on, somebody listening to the microphone can see the game, listen to your report, and start understanding that youÕre talking about the football or the wind in the stadium.Ó
Because of the similarity of human and monkey brains, the lab used monkeys fitted with cortical electrodes to study the neural signals - particularly those involved with motion. A monkey would be trained to place its finger on a point on a screen. Then, a target would appear while the monkey waited for a ÒgoÓ cue to move its finger there. During this delay period, the monkeyÕs brain would visualize the motion necessary for touching the target. The brain signals corresponding to neural activity during this process were recorded and compiled into a pattern that could be used to identify the signals responsible for this type of action.
The main characteristic of an examined neural signal is its average number of voltage spikes. ShenoyÕs team has developed a sort of Òcode bookÓ to translate what the neural signals are saying. For instance, a high average could indicate an arm moving to the right while a low average could be to the left. Such a code book would have to be made for each individual brain as a calibration.
A robust method for neural signal processing
In order to increase the capabilities of their brain-computer interface, NPSL has taken its own path in terms of signal processing. Shenoy revealed that they have systematically redesigned the way all the signal processing and the estimation algorithms worked for converting neural chatter in the brain into control signals. "We are able to show a quadruple performance improvement because of our new algorithms, above and beyond any invasive or non-invasive system in the past,Ó Shenoy comments. The ultimate goal in this area is to increase the rate of information flow which is currently up to 6.5 bits per second. ÒEvery single step in this process is rock solid mathematically. It is extremely quantitative," he says. "We donÕt get it perfectly; thereÕs some error, but on average we can get it right.Ó Once the computer knows the neural signalÕs message, it can convey it to the prosthetic for execution.
Future applications
The applications of a reliable brain-computer interface are vast. Amputees would be able to control their artificial limbs with their thoughts, just as they would move their natural limbs. In addition, new research being done on feedback nerve stimulation may one day provide proprioceptive sensation (the sense of the limbÕs spatial location) as well as a sense of touch for those with these advanced prosthetics.
Paralysis patients currently rely on a small repertoire of motions such as eye movement or breathing to control their machines. A brain-computer interface would allow them to perform much more complex tasks such as using a cursor on a computer to communicate or to operate a computer-controlled wheelchair. For now, Shenoy and his lab group are working to overcome the challenges of making neural prosthetics a technology, not of science fiction, but of the very real present.