Multiunit recording

Neighbouring Cells

Spike Sorting

Tetrodes
		[picture of tetrode]

Algorithm







What is Spike Sorting?

When an extracellular microelectrode is introduced into grey matter it
couples to a number of cell bodies and other action potential-carrying
membranes in the tissue.  The conventional procedure for recording
spikes extracellularly is to maneuver the electrode tip until the a
single membrane dominates the signal and thus record spikes from this
single source, a procedure commonly called "isolation".  More
recently, interest in the possibility of identifying distinct spike
events on multiple membranes without isolation has risen.  The idea is
that the current paths between the electrode and each spiking membrane
will be slightly different.  This, coupled with differences in the
intrinsic time course of the spike on different membranse, means that
waveforms from the different sources will be slightly different.  The
notion is that this variation in waveform can be exploited to
distinguish between the different sources.

This is quite difficult with a single electrode (although it has been
tried).  The tetrode is an attempt to make this problem easier by
recording currents from four tips distributed in a small volume of the
tissue.  Each tip couples to roughly the same group of membranes, so
the extra tips are not being used to increase the recording volume (as
is the case with many multi-electrodes), but rather to exploit stereo
effects to resolve different spikes in one region.  The statistical
problem of spike-sorting is quite similar using a single electrode or
a tetrode, although it is arguably only really tractable in the latter
case.


How do we approach the problem?

Historically, spike sorting has been carried out by eye.
Experimenters draw boundaries either around waveforms on an
oscilloscope-like display, or around clusters of points on a
computer-generated scatter-diagram of waveform features.  

We are developing a new, statistically well-founded technique for
"automatic" spike sorting.  The statistical solution to this problem
is important for at least two reasons: first, it ensures that the
separations achieved (and therefore results derived from them) are
reproducable and that the success of the separation is quantifiable;
second, it allows the recording technology to be scaled up so that
more signals than could reasonably be separated by hand can be
collected.

Our approach is based in explicit probabilistic modeling.  Where
possible, we measure the distribution of waveforms