Brain–Computer Interfaces - Index of

208 D.M. Taylor and M.E. Stetner
action potential. Therefore neurons with large cell bodies generate larger spikes in
the measured voltage than smaller neurons. However, the size of the spike is also
affected by how far away the neuron is from the electrode. The size of the voltage
spike drops off rapidly for neurons that are farther away. The end result is that these
microelectrodes often pick up a lot of small overlapping spikes from hundreds or
even a thousand different neurons located within a few microns from the recording
site. However, each microelectrode can also pick up larger spikes from just a few
very close neurons that have relatively large cell bodies.
The many small overlapping spikes from all the neurons in the vicinity of the
electrode mix together to form a general background activity level that can reflect
the overall amount of activation within very localized sections of the brain (much
like taking the local “temperature” in a region only a couple hundred microns in
diameter). This background noise picked up by the electrodes is often called “multiunit
activity” or “hash”. It can be very useful as a BCI control signal just like
the electric field potentials measured outside the brain can be a useful measure of
activity level (a.k.a. temperature) over much larger regions of the brain.
The real fun begins when you look at those few large nearby neurons whose
action potentials result in larger spikes in voltage that stand out above this background
level. Instead of just conveying overall activity levels, these individual neural
firing patterns can convey much more unique information about what a person is
sensing, thinking, or trying to do. For example, one neuron may increase its firing
rate when you are about to move your little finger to the right. Another neuron may
increase its firing rate when you are watching someone pinch their index finger and
thumb together. Still another neuron may only fire if you actually pinch your own
index finger and thumb together. So what happens when spikes from all these neurons
are picked up by the same electrode? Unless your BCI system can sort out
which spikes belonged to which neurons, the BCI decoder would not be able to tell
if you are moving your little finger or just watching someone make a pinch.
Signal processing methods have been developed to extract out which spikes
belong to which neuron using complicated and inherently flawed methodologies.
This process of extracting the activity of different neurons is depicted in Fig. 2 and
summarized here. Spikes generated from different neurons often have subtle differences
in the size and shape of their waveforms. Spikes can be “sorted” or designated
as originating from one neuron or another based on fine differences between their
waveforms. This spike sorting process is imperfect because sometimes the spike
waveforms from two neurons are too similar to tell apart, or there is too much noise
or variability in the voltage signal to identify consistent shapes in the waveforms
generated from each neuron.
In order to even see the fine details necessary for sorting each spike’s waveforms,
these voltage signals have to be recorded at a resolution two orders of
magnitude higher than the recording resolution normally used with non-penetrating
brain electrodes. Therefore, larger, more sophisticated processors are needed to
extract all the useful information that is available from arrays of implanted microelectrodes.
However, for wheelchair-mobile individuals, reducing the hardware and
power requirements is essential for practical use. For mobile applications, several