Life has gotten a bit easier lately for some Stanford lab monkeys.
Instead of earning their reward of fruit juice by using their arms to
chase blinking cursors on a computer screen, the primates in electrical
engineer Krishna Shenoy's lab merely have to think about approaching their targets.
That's because Shenoy's team is using tiny implanted electrodes to
decipher the brain signals that specify where the animals intend to
move their arms.
The ability to extract this ``plan activity'' is driving the latest
generation of ``neural prosthetics'' -- communication devices that
could one day help paraplegics by converting their cognitive signals
into electronic commands that move a computer cursor, for instance. A
patient could place phone calls using a keypad, type an e-mail message
or select menu options such as ``turn on the light'' or ``switch off
the TV.''
During the past decade, scientists have shown that such devices
work. But for the Stanford group, the bigger question is just how well
they can work.
``We're really trying to push the limits of how quickly and
accurately someone can communicate using a prosthetic device,'' said
Gopal Santhanam, an electrical engineering graduate student in Shenoy's
lab. With Stanford neurosurgeon Stephen Ryu, Santhanam presented the
team's findings Sunday at the annual meeting of the Society for
Neuroscience in San Diego.
The earliest neural prosthetics focused on restoring movement in
paralyzed patients by harnessing signals from the brain's motor cortex
to maneuver a mechanical arm or leg. But for communication prosthetic
systems where the goal matters more than the process of getting there
-- selecting the number ``4'' on a keypad, for instance -- scientists
realized they didn't need to bother decoding the instructions for
continuous movement.
Rather than translate the motor signals for ``move my hand in a
particular direction,'' they began to home in on regions of the brain
that begin firing before movement begins. Within these so-called
planning areas, researchers could tap into the neurons that shout, ``I
want to touch that 4 key.''
``It's not representing the motor parameters of the movement but
rather the intention of the subject,'' said Richard Andersen, a
neuroscientist at the California Institute of Technology, whose work in
plan-based prosthetic systems was published in Science magazine in
July. Shenoy was a postdoctoral fellow in the Andersen group before
starting his own lab at Stanford in 2001.
To develop their prosthetic device, Shenoy's team devised a scheme
to get the monkeys to behave like paraplegics -- using their thoughts
alone to select a computer cursor occupying one of eight different
positions. As a first step, the monkeys were trained to reach for new
targets once they appeared on the screen. Every correct touch earned a
drop of juice.
``Eavesdropping'' on the brain's planning regions, the researchers
were able to identify signature patterns of nerve cell activity
corresponding to each of the eight cursor positions. They created
algorithms to extract and decode this plan activity during the split
second before the monkey prepared to reach for its target.
If the scientists correctly interpreted these plan signals, the
animal got a drop of juice and the next cursor would appear. ``Because
the monkey is rewarded, he just realizes there's no point in making the
reach,'' Santhanam said.
The system can make up to 3.6 selections per second on a computer
screen when two possible targets are presented, and it did this with
more than 94 percent accuracy, the researchers said.
Acknowledging that it can be difficult to quantify and compare how
well these devices work, Shenoy estimates that his team's device can
achieve nine to 10 words per minute -- about a threefold increase over
previous systems.
``It's a field where an extra word per minute is substantial,''
Shenoy said. ``We're desensitized to it in Silicon Valley, where the
doubling of a processor's speed is sort of expected. It's not a given
in this field.''
But neural prosthetics have not even begun to approach the
complexity of human cognition, others caution. ``This could be a
lifesaver for a paraplegic. To be able to move something is
wonderful,'' said Jeff Hawkins, author of ``On Intelligence,'' a new
book about brain-based machines, and founder of Redwood Neuroscience
Institute in Menlo Park. ``But even when you're making four decisions a
second, that's tiny compared to the normal capacity of the brain.''
Still, Shenoy approaches the challenge with an engineer's intuition
-- likening the brain-to-computer link in neural prostheses to data
transmission in telecommunications. ``For a long time, we lived with
our 56K modems,'' he said. ``If we're smarter about our signal
processing, you can transmit a lot more.''