How Brain-operated Machines Can Be Stable, Functional
DURHAM, N.C. -- In order to function stably over long
periods, brain-operated devices such as neural prosthetic limbs
for paralyzed people will require brain signals fed from
hundreds of infinitesimal recording electrodes in the brain,
Duke University researchers have concluded.
Their findings in studies with monkeys are defining the
requirements for successful brain-machine systems, as the
researchers progress toward the first clinical trials of fully
functional neural prosthetics.
The researchers, led by neurobiologist Miguel Nicolelis,
published their findings in the Nov. 16, 2005, issue of the
Journal of Neuroscience. Besides Nicolelis, co-authors of the
study were Jose Carmena and Mikhail Lebedev in the Nicolelis
laboratory and biomedical engineer Craig Henriquez in the Pratt
School pf Engineering. All are members of the Duke Center for
Neuroengineering. The research was sponsored by the Christopher
Reeve Paralysis Foundation, the Defense Advanced Research
Projects Agency, the National Institutes of Health, the
National Science Foundation and the James S. McDonnell
Basically, the researchers gave macaque monkeys a task
involving reaching with a hand-held pole and analyzed the
stability of the outputs from neurons in the monkeys' brains,
as measured by scores of recording electrodes.
They found that individual electrodes, or even small
numbers, showed variability in output that would render
brain-operated devices inconsistent in their operation.
However, the researchers found that integrating signals from
large ensembles of neurons yielded a stable output that could
support complex functionality of neural prosthetics such as
arms and hands.
"Ever since the initial discoveries that brain signals could
operate external devices, there has been a debate over the
neuronal output required to operate such devices," said
Nicolelis. "It is technically very difficult to record from
many neurons and integrate the signals, so some in the field
have advocated using smaller samples of neurons to operate
"We have held that the number should be in the hundreds,
while other groups have proposed that they could get by
recording from ten or twenty neurons," he said. "This paper
shows that such numbers would not work. A clinically relevant
device --such as an arm that would produce continuous movement
over many years -- would require sampling from much, much
larger numbers of neurons than what has been proposed by other
"Output from many cells is necessary, because there is
fluctuation in output among individual neurons," said
Nicolelis. "At a particular moment, some cells might give quite
good output, while at other times, they would not be adequate.
It's like taking a poll before an election. The question is how
many people you need to take a reliable poll of who will win,"
"Our studies show the brain is continuously sampling from
large ensembles of neurons to achieve a stable output, such as
controlling a limb" said Nicolelis. "It samples from such a
huge number of cells to enable it to remove noise and average
the contributions of the network to achieve such
According to Nicolelis, evolution has favored such
redundancy, since it enables the brain to lose cells or to
tolerate inconsistency in individual cells and still maintain a
Similarly, he said, more extensive electrode implants for
neural prosthetics would make clinical sense because it would
avoid subsequent surgeries necessary to re-implant electrodes
if any malfunctioned, or if the cells became
"The technology already exists for such large electrode
arrays, and the surgery need not be any more complex than
implanting a heart pacemaker," said Nicolelis.
In their experiments, the researchers gave macaque monkeys a
reaching task, in which the animals had to operate a
joystick-like control to move a cursor over a target, in order
to receive a juice reward. The researchers had implanted arrays
of up to 64 hair-thin recording electrodes in the animals'
In their analysis, the researchers systematically varied the
number of neurons that they included in integrating to provide
the output, and measured the impact on output stability. When
few neurons were involved, they found, the output was unstable;
but large ensembles of neurons yielded stable output.
In further studies, Nicolelis and his colleagues are
exploring whether they can develop techniques to extract from
the same ensemble of neurons multiple signals for controlling
different components of a device -- such as different "muscles"
in a robotic arm. It may be that such multiplicity of outputs
can arise from the same ensemble, because different
subpopulations of neurons contribute to different outputs, said