Velocity-selective Recording using Multi-electrode Nerve Cuffs
|
Nick Donaldson |
John Taylor and Jeff Winter |
|
Dept. of Medical Physics
& Bioengineering University College London 11-20 Capper St, London,
WC1E 6JA, UK nickd@medphys.ucl.ac.uk |
Dept. of Electronic
Engineering University College London Torrington Place, London,
WC1E 7JE, UK |
Hansen et al. [1] have shown that signals from cuff electrodes can be used
chronically as inputs to neuroprostheses. This paper describes methods (a) for
improving the signal-to-noise ratio (S/N) of neural signals picked up from
cuffs and, (b) for discriminating between activities in different groups of
neurons according to their propagation velocities.
Jezernik et al. [2] used signal processing of the
signal from a tripolar cuff for these purposes: Wiener filters improved S/N;
and the signal was classified according to activity in three types of receptor
by autocorrelation and an artificial neural network. By contrast, our method
employs a multi-electrode cuff to provide several tripolar signals, which are
then subjected to relatively simple signal processing operations.
Pflaum et al. [3] introduced an amplifier configuration for tripolar cuffs
in which two front-end amplifiers are connected to the three electrodes (“true
tripole”). That configuration is shown extended in Figure 1 for N+2 electrodes
and N 2nd-rank amplifiers. If one action potential (AP) propagates through the
cuff, it produces the same response at the outputs of all the 2nd-rank
amplifiers, staggered in time by the electrode pitch divided by the velocity (T,
2T, 3T.. NT). Since T is velocity dependent, the artificial
time delays (Nt...3t, 2t, t) will be matched to one propagation velocity. If
they are matched, the AP responses are temporally re-aligned so that the output
of the adder is greater than all the individual 2nd-rank amplifier responses by
a factor N.
Assume that the length of
the cuff is determined by anatomical and surgical considerations. The Riso
& Pflaum true-tripole has three
electrodes. As the number of electrodes is increased and hence the electrode
pitch decreases, the signal amplitude at the outputs of each 2nd-rank amplifier
decreases, but more signals are added together. Our theory predicts that there
is a maximum in overall amplitude for a particular pitch corresponding to an
optimal number of electrodes for a given cuff length [4].
Four types of noise are
present: Johnson noise from the electrodes’ access resistances; Johnson noise
from the axial resistances inside the cuff; amplifier voltage noise; and
amplifier current noise. Analysis shows that when all these are appropriately
summed, the total rises at a rate that is slightly less than ÖN.
For example, we considered a 30mm-long cuff. The maximal summed AP signal, for fibres of diameters 5, 10, 15 & 20 mm, occurs with 9 to 11 electrodes at a pitch of 3.75 to 3.00 mm. After the adder, the combined signal is 5 to 8 times larger, depending on fibre diameter, than from a single tripole (i.e. pitch 15 mm). For 11 electrodes, and assuming reasonable values for the resistances and amplifier noise characteristics, the predicted noise increases by a factor of 2.2 so the improvement in S/N is 2.3 to 3.6 (20 to 5 mm diameter, respectively).
Figure 1 shows functional
blocks on several planes after the 2nd-rank amplifiers. Each plane has one
output and time delays (Nt .…. 3t, 2t, t) that are matched to one
propagation velocity, which may be positive or negative (afferent or efferent).
On each plane there is also a narrow bandpass filter. A crucial part of this
method is this: that the centre frequency of the passbands of these filters is
proportionately related to the matched velocity. In a particular system with
electrode pitch d and matched
velocities (v1, v2, v3 ...vp), the
artificial time delays for planes 1 to p
will be t1=d/v1, t2=d/v2, t3=d/v3 ... tp=d/vp and the centre
frequencies will be f1=v1/l, f2=v2/l, f3=v3/l....fp=vp/l, where l is a constant length which
is be chosen by the designer. This proportionality means that the Q of the velocity-selective filter in
the velocity domain is the same for all planes, which makes the filter outputs
much easier to interpret. So if the delays (tp) in Fig 1 are
chosen appropriately to cover the required velocity spectrum (probably in
overlapping bands) and, in addition, the centre frequency of each
Figure 2: Outputs of three Velocity-Selective Filters
bandpass filter is chosen
according to the above simple rule, then the relative activity in each velocity
band can be measured directly.
We expect that this method
will allow velocity-selective recording and
improved S/N in nerve cuff recordings. Activity in equal-sized fibres but
propagating in opposite directions should be distinguishable, as should
activity in different-sized fibres propagating in the same direction. To be
practicable, we expect that the amplifiers must be mounted on the cuff, and
this is an objective of the SENS Project (EU Grant QLG5-CT-2000-01372).
The invention described above is subject of a Patent Application (British Patent Office, Number 0201100.5, filed 18th January, 2002).
[1] Hansen M., Haugland M., Sinkjaer T & Donaldson N. (2002) “Real time drop foot correction using machine learning and natural sensors.” Neuromodulation, 5, 41-53.
[2] Jezernik S., Grill W. & Sinkjaer T. (1999) “Neurographic recordings, electrical stimulation and new neural signal processing methods for closed-loop neuroprosthetic control of bladder hyper-reflexia.” Proc. IFESS Conf., 81-84.
[3] Pflaum C., Riso R. & Wiesspeiner (1995) “An improved nerve cuff recordings configuration for FES feedback control systems that utilise natural sensors.” Proc. IFESS Conf., 407-10.
[4] Winter J., Rahal M., Taylor J., Donaldson N. de N. & Struijk J.J. (2000) “Improved spatial filtering of ENG signals using multielectrode nerve cuff.”Proc IFESS Conf., 368-71