Medical Engineering & Physics 31 (2009) 945951

Contents lists available at ScienceDirect

Medical Engineering & Physics

journal homepage: www.elsevier.com/locate/medengphy

Array electrode design for transcutaneous electrical stimulation:

A simulation study
Andreas Kuhn a,d, , Thierry Keller a,b , Silvestro Micera a,c , Manfred Morari a
a

Automatic Control Laboratory, ETH Zurich, Switzerland

Biorobotics Department, Fatronik-Tecnalia, Donostia-San Sebastian, Spain
ARTS Lab, Scuola Superiore SantAnna, Pisa, Italy
d
Altran AG, Technology and R&D Consulting, Zurich, Switzerland
b
c

a r t i c l e

i n f o

Article history:
Received 18 October 2008
Received in revised form 29 April 2009
Accepted 1 May 2009
Keywords:
Transcutaneous electrical stimulation
Array electrode design
Finite element model
Gap size
Skin interface layer resistivity

a b s t r a c t
Array electrodes are a promising technology that is likely to bring transcutaneous electrical stimulation
(TES) a step forward. The dynamic adaptation of electrode size and position helps to simplify the use of
electrical stimulation systems and to increase their clinical efcacy. However, up to now array electrodes
were built by trial and error and it is unclear how, for example, the gaps between the array elements or
the resistivity of the electrodeskin interface material inuence the current distribution. A TES model
that comprises a nite element model and a nerve model has been used to analyze the inuence of array
electrode gaps and gel resistivities on nerve activation. Simulation results indicate that the resistivity of
the electrodeskin interface layer should be adapted depending on the size of the gaps between the array
elements. Furthermore, the gap sizes should be smaller than 3 mm in order to keep losses small.
2009 IPEM. Published by Elsevier Ltd. All rights reserved.

1. Introduction
In transcutaneous electrical stimulation (TES) pairs of surface
electrodes are placed on the skin in order to stimulate motor nerves.
Clinically, TES is often applied in the rehabilitation of stroke subjects
or spinal cord injured subjects [1], or for supporting tasks of daily
living [2] using so called neuroprostheses. In the past such neuroprostheses used single stimulation electrodes [3,4]. Recently array
electrodes were proposed to improve the efcacy of such TES systems [5,6]. Array electrodes consist of multiple elements which can
be individually activated to form a virtual electrode of arbitrary size
and location. The position and size of the activated region (virtual
electrode) can be dynamically changed [7,8]. For good wearability
textile array electrodes are produced by embroidering silver coated
bers in the form of array electrodes into garments [9]. One large
layer of self-adhesive hydrogel is used as skin interface between the
skin and the array electrode [10]. The gaps between array elements
and the resistivity of the interface layer (gel) inuence the current
distribution owing into the human limb.
In previous works the sizes of the gaps between array elements
and the resistivity of the gel layer were chosen intuitively [5,6,10].
In this paper a TES model comprising a nite element (FE) and a

Corresponding author at: Automatic Control Laboratory, ETH Zurich, Switzerland. Tel.: +41 44 632 6571.
E-mail address: kuhnan@control.ee.ethz.ch (A. Kuhn).

nerve model is used in order to analyze and better understand the

inuence of the gap sizes and the gel resistivity on nerve activation.
Using the TES model it was investigated how large the gaps are
allowed to be and how this decision is related to the choice of the
gel resistivity.
Previously, the indifferent electrode (anode) was placed separately from the array [5,6]. However, in [9] it was shown that the
active (cathode) and the indifferent electrode (anode) can be placed
on the same array. This simplies the application of TES because
only one electrode has to be applied to the human body instead
of one array plus a separate indifferent electrode. Obviously, a part
of the applied current is lost because it will directly ow from the
cathode to the anode through the hydrogel. It is unclear how much
of the applied current is lost in the gel layer depending on different parameters (e.g., gel resistivity, skin resistivity, . . .). To address
these issues different parameter combinations were applied to the
TES model in order to quantify and to reduce these losses.
2. Methods
2.1. Combined FE and nerve model
The experimentally veried TES model comprised a FE model
and a nerve model [11]. The FE model calculates the 3D electric
potential distribution in a geometry representing the forearm and
the nerve model relates the potential distribution to nerve activation [12].

1350-4533/$ see front matter 2009 IPEM. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.medengphy.2009.05.006

946

A. Kuhn et al. / Medical Engineering & Physics 31 (2009) 945951

Fig. 2. (A) Two single electrodes for comparison with array electrodes. Grey electrodes indicate anodes and electrode with crossed lines cathodes. (B) Array electrode
with separate single anode. (CF) Array with different distances between anode and
cathode (C: 51 mm, D: 35 mm, E: 19 mm, F: 3 mm).

identify parameters where activation of array electrodes is similar to the activation of single electrodes. Therefore, two different
electrode layouts were considered:
An array electrode and a single electrode were placed on the
cylinder (sketches B to F in Fig. 2).
Two single electrodes were used (sketch A in Fig. 2).

Fig. 1. Cylindrical arm model (length: 60 cm, diameter: 10 cm) with array electrode
and a separate single electrode. Underneath the array one large gel layer provides
the interface between array elements and skin. A region of 2 2 array elements was
used as cathode. The anode was either on the separate single electrode or on the
array.

The electric scalar potential distribution (VFE ) in the arm model

generated during TES by array electrodes was calculated by using
Laplaces Eq. (1) with the FEM software Ansys (EMAG, Ansys Inc.,
Canonsburg, PA). It takes only the resistive properties () into
account. Capacitive and inductive effects were neglected in the
model [13,12]. The geometry was a multiple layer cylindrical representation of the forearm consisting of skin, fat, muscle and bone
layers (Fig. 1). The resistivity of the skin (skin ) depends non-linearly
on the applied current density [14]. The current densities that are
typically applied are between 1 mA/cm2 and 15 mA/cm2 [9] leading to skin resistivities between 670  m and 2100  m [14]. In
order to cover a wide range of these experimentally found skin
resistivities, low (700  m) and high skin resistivities (2000  m)
were used in the TES model. The thickness of the skin in the model
was 1.5 mm. The thicknesses and electrical properties of the other
tissues were 2.5 mm and 33  m for fat, 33.5 mm and 3  m for muscle, 6 mm and 50  m for cortical bone, and 6.5 mm and 12.5  m
for bone marrow [15,16]. The total cylinder diameter is therefore
10 cm. The anisotropy of the muscles resistivity was considered
by using a factor of three between the axial and radial direction
(3axial = radial ) [16]. The cylinder was chosen long enough (60 cm)
that the boundary of the cylinder did not have an inuence on the
potential distribution underneath the electrodes.
([] VFE ) = 0

(1)

The TES model was used to compare array electrodes with standard single electrodes. The inuence of gaps between individual
array elements and the inuence of the electrical properties of the
electrode gel on nerve activation were investigated. The goal is to

Between cathode and anode biphasic current regulated pulses

with a pulse amplitude I and a pulse duration of 300 s were used.
The single elements of the array electrode were modeled as good
conducting substrate (silver) and had a size of 13 mm (lelem in Fig. 1),
the gap sizes (lgap ) were between 1 mm and 5 mm (lgap in Fig. 1), and
the resistivity was 1.6 108  m (silver). These parameters were
chosen according to the specications of array electrodes used in
another study [9]. Underneath the array elements a single hydrogel
layer provided an interface to the skin. This gel layer had thicknesses
(lgel ) between 0.25 mm and 2 mm, and resistivities (gel ) between
1  m and 10,000  m. This is the range of hydrogels that can be
produced by manufactures such as Sekisui Plastics Co., Ltd., Tokyo.
In order to compare single with array electrodes their edge
lengths have to be of comparable size. The size s of the single
cathode (sketch A in Fig. 2) was always adapted to the size of the
corresponding 2 by 2 electrode (array, sketches B to F in Fig. 2)
it was compared with. The edge length of the single cathode was
s = 2lelem + lgap . The centers of the single electrode and the 2 2
virtual electrode were kept congruent.
Passive nerve bundle models consisting of 100 axons with diameters distributed according to the bimodal distribution in human
nerve bundles with peaks at 6 m and 13 m [17,18] were placed
within the volume representing muscle and were oriented in axial
direction of the cylinder. Hundred axons are enough to accurately
predict recruitment of one nerve bundle [19]. The nerve bundles
contained passive axon models that were derived from the MRGmodel [20]. It was previously shown that such passive axon models
can be used to predict activation of active axon models [21,22]. Computer simulations of the MRG model were performed in NEURON
(Yale University). An implementation of a single axon can be downloaded from the model database (ModelDB) on the NEURON web
page (http://www.neuron.yale.edu/neuron). Multiple of these single axon models with the mentioned bimodal diameter distribution
were used in order to model a nerve bundle containing axons of
different diameters. The axons intracellular voltage is inuenced
by the extracellular potential Ve,n at the nodes of Ranvier (n) and
at non-nodal compartments (n n) between the nodes of Ranvier
Ve,nn (t) [20].
The link between the FE model and the nerve models was established by assigning the spatially interpolated potentials from the FE
model VFE,n and VFE,nn to the corresponding extracellular potentials of the axon models Ve,n = VFE,n and Ve,nn = VFE,nn .

A. Kuhn et al. / Medical Engineering & Physics 31 (2009) 945951

947

2.3. Losses in the gel layer when anode and cathode are on the
same array

Fig. 3. Supercial and deep nerve within the volume representing muscle tissue. The
locations indicated with nd are the places where nerve activation was investigated
and represent common extreme nerve depths found in humans [23].

Nerve activation in the TES model was assessed underneath the

cathode (active electrode). The worst activation using the array
electrode compared to using the single electrode is underneath the
gap in the center of the 2 2 active region (nd in Fig. 2). Because
array electrodes should produce similar activation to single electrodes this location of worst activation was investigated for two
nerve depths (d); a supercial (1 mm) and a deep (16 mm) nerve as
depicted in Fig. 3. These two nerve depth represent extreme nerve
depths commonly found in humans [23].
2.2. Inuence of gel resistivity and array gaps on nerve activation
The inuence of the gaps between the single array elements on
nerve activation was investigated. The aim was to nd out what gel
resistivities should be used for different gap sizes such that nerve
activation underneath the gaps is similar to that with single electrodes. For that purpose activation generated using array geometry
B (Fig. 2) was compared with the activation generated using single
electrode geometry A (Fig. 2). The inuence was quantied by calculating the additional current Iadd that was required in the array
geometry to generate the same nerve activation at locations nd as in
the single electrode geometry. The additional current Iadd was calculated for the two nerve depths (d) of 1 mm and 16 mm shown in
Fig. 3 and for an intermediate recruitment (11 m axon diameter).
When axons down to a diameter of 11 m are activated in a nerve
bundle an intermediate recruitment of about 50% is present [16].
Apart from the worst case (nerve activation underneath the
gaps) the mean additional current Iadd,area at multiple locations on
an area underneath the active electrodes was calculated. The area
was located on cylinders that were concentric with the arm model.
One of the areas on which the mean of the nerve activation was
calculated is indicated with Area (P) in Fig. 1. Two cylinder radii (corresponding to nerve depths 1 mm and 16 mm) were used in order to
calculate Iadd,area at different depths. The same two physiologically
motivated nerve depths as in Fig. 3 representing a supercial and a
deep nerve were used.

The losses that occur when the cathode and the anode are on
the same array were investigated. Depending mainly on parameters
like gel resistivity gel , gel layer thickness lgel , skin resistivity skin ,
and the distance between cathode and anode, a certain part of the
applied current ows directly through the gel layer from cathode
to anode. Our goal was to nd out how high the gel resistivity gel
should be in order to keep these losses small. The losses were quantied by calculating the additional current Iloss that was required
to generate the same nerve activation using geometries C, D, E or
F (Fig. 2) compared to geometry B (separate anode). Because the
losses are larger for deeper nerves this additional current Iloss was
calculated for a deep nerve at 16 mm nerve depth (Fig. 3) and an
intermediate nerve recruitment (11 m axon diameter).
3. Results
3.1. Inuence of gel resistivity and array gaps on nerve activation
Qualitatively, the inuence of gel resistivity on electrical current distribution is shown in Fig. 4. The upper sketch shows the
spreading of the current for a high resistivity (e.g., 10,000 m) of
the gel layer and the lower sketch for a low resistivity (e.g., 1  m).
A higher gel resistivity focuses the current underneath the single
pads. As a result supercial nerves in region A are activated less
compared to region D and deep nerves in region B are more activated compared to region E. In adjacent regions to the activated
array elements activation is lower for high resistivities (region C)
than for low resistivities (region F).
Quantitatively, the additional current Iadd that is required to
achieve the same activation with array electrodes compared to single electrodes is plotted in Fig. 5 versus the gel resistivity gel for
different gap sizes lgap and nerve depths d as described in Section
2.2.
It can be seen by comparing Fig. 5(a) with Fig. 5(b) that the
inuence of the gap size lgap and of the gel resistivity gel on Iadd
is not the same for supercial and deep nerves. The curves for the
different gap sizes in Fig. 5(a) (supercial nerve) are further apart
compared to the chart in Fig. 5(b) (deep nerve). This indicates that
the gap size lgap has a larger inuence on supercial nerves compared to deep nerves because the current spreads within the muscle
and diminishes the inuence of the array gaps on deep nerves. For
low gel resistivities (gel < 20  m) Iadd is large for both supercial
and deep nerves. The reason is that the current spreads in the gel
layer and the effective electrode is much larger than the 2 2 active
elements on the array. For large gel resistivities (gel > 3000 m)
Iadd is not the same for supercial and deep nerves. Due to the

Fig. 4. Depicts qualitatively (left) and quantitatively (right) the inuence of different gel resistivities on the spreading of the current. In the simulation results from the FE
model (right) the normalized current density is depicted (skin = 700  m, current density on 2 2 array is 5 mA/cm2 ). Both upper images show the spreading of the current
for a high gel resistivity (10,000  m) and the lower images for a low gel resistivity (10  m). It can be seen that with a higher gel resistivity the current is more focused under
the individual array elements.

948

A. Kuhn et al. / Medical Engineering & Physics 31 (2009) 945951

Fig. 5. The additionally required current Iadd versus the gel resistivity are shown for different parameter combinations. Iadd was calculated comparing geometry B (Fig. 2) and
geometry A. (a) Iadd for a supercial nerve (1 mm nerve depth). (b) Iadd for a deep nerve (16 mm nerve depth).

Fig. 6. Mean of Iadd,superficial (Fig. 5(a)) and Iadd,deep (Fig. 5(b)) (denoted with Iadd ) versus the gel resistivity are shown for different gap sizes. (a) Iadd for a skin resistivity of
700  m. (b) Iadd for a skin resistivity of 2000  m.

small spread of the current at high gel resistivities high currents

are required to activate supercial nerves underneath large array
gaps (lgap > 3.5 mm). For deeper nerves this effect is diminished
due to the spreading of the current in the muscle.
The curves depicted in Fig. 5 allow to choose a gel resistivity so
that the additional current Iadd is minimal depending on the gap size
lgap of the array. For example, a gel resistivity of 100  m should be
used when the gap size is 5 mm and a supercial nerve is stimulated
(Fig. 5(a)). However, the results in Fig. 5 show that these optimal
gel resistivities are not the same for supercial and deep nerves. To
equally weight supercial and deep nerves the mean of the curves
in Fig. 5(a) (supercial) and Fig. 5(b) (deep) were calculated. These
curves are shown for a skin resistivity of 700  m in Fig. 6(a) and
for a skin resistivity of 2000  m in Fig. 6(b). The minimal values
for the different curves in both plots of Fig. 6 are plotted in Fig. 7.
Larger gap sizes require a larger spread of the current in the gel
layer and therefore a smaller gel resistivity (Fig. 7). Furthermore,
the results indicate that for a xed gap size larger skin resistivities
require larger gel resistivities.
The additional current Iadd at the optimal gel resistivity for the
different gap sizes (minima of each curve) in Fig. 6 is larger for larger
gap sizes (e.g., Iadd is 5 mA for a gap size of 5 mm but only 1 mA for
a gap size of 1 mm). This shows that lower losses can be achieved

Fig. 7. The plot shows the preferred (in terms of losses) gel resistivities for different
gap sizes of the array. The values were derived from the minima of the curves in
Fig. 6.

A. Kuhn et al. / Medical Engineering & Physics 31 (2009) 945951

949

Table 1
Mean additional current Iadd,area for different gap sizes and nerve depths. The threshold current is the current required on a single electrode (adapted size s = 2lelem + lgap
described in Methods) for 50% recruitment. The skin resistivity skin was 2000 m
and the gel resistivity gel was 3000 m.
Gap size lgap
[mm]

Nerve depth
d [mm]

Mean additional
current Iadd,area [mA]

Threshold current of
single electrode [mA]

1
2
3
4
5
1
2
3
4
5

1
1
1
1
1
16
16
16
16
16

0.4
0.6
0.8
1.2
1.6
1.6
1.8
2
3.2
4

2.5
2.6
2.8
3.1
3.2
27.8
28.9
29.9
30.9
32.8

with smaller gap sizes. The results in Fig. 6 indicate that in order
to keep the additional currents Iadd below 2 mA, gap sizes below
3 mm are required.
Apart from the worst case (nerve underneath the gaps) the mean
additional current Iadd,area underneath the array was calculated (see
Section 2.2). The losses are much lower (Table 1) than the worst
cases in Fig. 5. For all tested gap sizes and nerve depths the losses
were 4 mA or less (Table 1).
3.2. Losses in the gel layer when anode and cathode are on the
same array
The losses Iloss versus the gel resistivity gel were calculated for
different cathodeanode distances, skin resistivities skin (Fig. 8),
and gel thicknesses lgel (Fig. 9) as described in Section 2.3.
For lower gel resistivities gel the losses Iloss increased because
a larger part of the current owed directly through the gel layer
and did not add to nerve activation (Figs. 8 and 9). A limit for
Iloss of 2 mA was introduced in order to obtain lower limits for the
gel resistivities where losses are at an acceptable level. Losses of
2 mA are acceptable because the additional current does not introduce much more pain [24] and does not require much more power
of the stimulator. For short cathodeanode distances (3 mm) the
losses were above 2 mA (Fig. 8). Therefore, cathodeanode distances

Fig. 9. The losses Iloss versus the gel resistivity are shown for different gel thicknesses.
The nerve depth is 16 mm and the gap sizes are 3 mm.

below 19 mm should be avoided. For cathodeanode distances

above 19 mm the losses Iloss were below 2 mA for gel resistivities
gel above 50  m.
The larger skin resistivity skin of 2000 m required a slightly
larger gel resistivity gel to achieve the same losses Iloss (Fig. 8).
However, compared to the cathodeanode distance and the gel
resistivity, the skin resistivity has a small inuence on the losses.
The results in Fig. 9 show that thicker gel layers produce larger
losses and that the inuence of the gel thicknesses on the losses
is larger at low gel resistivities. However, because gel resistivities
above 50  m should be used as described in the two previous paragraphs the results show that only a little gain can be achieved with
thinner gel layers.
All in all, these results show that for cathodeanode distances
above 19 mm and gel resistivities gel above 50  m the losses are
below 2 mA for all tested combinations of parameters (Fig. 8). This
denes a lower limit for the cathodeanode distance and the gel
resistivity gel when cathode and anode are placed on the same
array.
4. Discussion

Fig. 8. The losses Iloss versus the gel resistivity are shown for different cathodeanode
distances and skin resistivities. The nerve depth is 16 mm and the gap sizes are 3 mm.

The array technology is a versatile approach that can dynamically change the size and position of the active region. This will
enable clinicians to reduce the time-consuming procedure of nding optimal electrode positions and sizes. However, it was unclear
how the gaps between the array elements and the resistivity of the
interface layer (gel resistivity) inuence activation. Furthermore,
it was unclear how large the losses are when cathode and anode
are placed on the same array connected through the gel layer. A
combined FE and nerve model was employed to clarify what gel
resistivities and gap sizes should be used. A modeling approach was
favored over experiments because it has the advantage that specic inuences can be investigated. With experiments the number
of different congurations that can be investigated is limited.
A constant array element size was used to calculate the presented simulation results. The sensitivity against changes of this
electrode size was checked for skin resistivities of 700  m and
2000  m and for the used range of gel resistivities. The inuence on the additional current Iadd and the mean additional current
Iadd,area at both nerve depths was below 5% for element edge lengths
between 0.9 cm and 2.3 cm.
The presented FE model neglects the electrochemical aspects of
the electrode/gel and the gel/skin interfaces. It was shown in an

950

A. Kuhn et al. / Medical Engineering & Physics 31 (2009) 945951

earlier study that these effects, which are mainly capacitive, can
be neglected when using current regulated TES [11]. The inuence
of the capacitance (permittivities r ) of the electrode interfaces on
nerve activation was below 0.1%. The reason is that the voltage is
regulated such that approximately a constant current is present at
the nerve site. However, for studies where not only nerve activation
but also comfort and the efcacy of the electrodes is of interest the
electrochemical aspects should be considered in the FE model.
A passive nerve model was used to describe nerve activation
instead of an active (non-linear) nerve model. Such passive models
can be used to describe nerve activation of rectangular pulses under
the cathode in a computationally efcient way [21,22]. However,
arbitrary current waveforms (e.g. exponential, sine, multi-pulse)
cannot be described. Also blocking of nerves by hyperpolarizing the
axons membrane or anode-break excitation cannot be described.
For these more complex activation schemes the non-linear ionic
currents have to be taken into account as for example shown in
[11,20].
The presented simulation results for gap sizes between 1 mm
and 5 mm (Section 3.1) indicate that low resistivities lead to a large
spread of the current within the gel layer. As a consequence the area
where the current enters the skin is larger than the single array
elements on the array. The advantage of the large spread is that
activation is uniform and that gaps have less inuence but unfortunately the activation is not selective any more. On the other hand the
results show that a too high a gel resistivity dramatically reduces
the activation underneath the gaps of the array because the current ows perpendicular down into the muscle and does not reach
the nerves below the gaps. As a consequence for high gel resistivities the activation is non-uniform but selective. For these reasons
the gel resistivity is an important design parameter that has to be
properly chosen depending on the gap sizes of the constructed array
electrode (Fig. 7) in order to achieve a good tradeoff between a uniform and a selective activation. Because the skin resistivity of most
humans is between 700  m and 2000  m [16,14] the gel resistivity
should be chosen according to the two curves depicted in Fig. 7.
Apart from the nerve activation underneath the gaps of the array
also the mean losses under the whole electrode were calculated.
These losses are smaller compared to the losses underneath the
gaps (Table 1). This gives evidence that only few places have a lower
activation (primarily underneath the gaps). However, for automatic
identication algorithms that nd optimal stimulation regions [25]
it is important that all locations underneath the electrodes experience similar activations. Therefore, it is important to produce a
uniform current distribution underneath the array.
The simulation results where cathode and anode were on the
same gel layer (Section 3.2) showed that the gel resistivity should
be at least 50  m and the cathodeanode distances should be more
than 19 mm in order to keep the losses below 2 mA. A gel resistivity
above 50  m is not restrictive when compared to the loss related
requirements in Fig. 7. Hence it is sufcient to ensure that cathode
and anode are never closer than 20 mm in order to keep losses
small when anode and cathode are on the same interface layer.
A thinner gel layer thickness decreases the losses when anode
and cathode are on the same gel layer (Fig. 9). However, the effect of
using a larger gel resistivity was found to be more effective (Fig. 9)
compared to a thinner gel layer thickness. Additionally the gel layer
thickness should not be thinner than 1 mm for safety reasons. When
using current regulated stimulation a break through of the interface
material can cause burns of the skin.
The presented results were calculated for parameters of a common human forearm. If an array electrode with gel resistivities from
Fig. 7 is applied to another body part the losses will not be optimally low. The main parameters that change when moving to other
body parts are the nerve depth and the fat layer thickness. A thicker
fat layer has a comparable effect as a deeper nerve because the

nerves are also further away from the stimulating electrode. The
effect of a deeper nerve can be estimated by comparing Fig. 5(a)
with Fig. 6(a). In the results of Fig. 6(a), both supercial and deep
nerves are taken into account whereas in Fig. 5(a) only the supercial nerves are taken into account. The optimal gel resistivities
were moved to higher values when incorporating deeper nerves.
However, the changes (e.g., for a gap size of 3 mm from 400  m to
1000  m) were small compared to the whole range of possible gel
resistivities for different gap sizes given in Fig. 7. This indicates that
the gap size is the dominant parameter and that other parameters
and therefore also different locations have a minor inuence on the
optimal gel resistivity.
In conclusion, for a good trade-off we suggest, based on the simulation results, that gap sizes should be smaller than 3 mm in order
to have losses less than 2 mA. For 3 mm gap sizes the preferred gel
resistivity is 1500  m, for 2 mm gap sizes 3000  m, and for
1 mm gap sizes 4500  m.
Acknowledgment
The project was supported by the Swiss National Science Foundation (SNF) No. 205321-107904/1.
Conict of interest
All authors conrm that there is no conict of interest.
References
[1] Baker LL, McNeal DR, Benton LA, Bowman BR, Waters RL. Neuromuscular electrical stimulation. 4th ed; 2000.
[2] Shefer LR, Chae J. Neuromuscular electrical stimulation in neurorehabilitation.
Muscle Nerve 2007;35(5):56290.
[3] Popovic MR, Popovic DB, Keller T. Neuroprostheses for grasping. Neurol Res
2002;24(5):44352.
[4] Snoek GJ, Jzerman IMJ, intGroen FA, Stoffers TS, Zilvold G. Use of the ness
handmaster to restore handfunction in tetraplegia: clinical experiences in ten
patients. Spinal Cord 2000;38(4):2449.
[5] Popovic-Bijelic A, Bijelic G, Jorgovanovic N, Bojanic D, Popovic MB, Popovic DB.
Multi-eld surface electrode for selective electrical stimulation. Artif Organs
2005;29(6):44852.
[6] Elsaify A, Fothergill JC, Peasgood W. A portable fes system incorporating an
electrode array and feedback sensors. In: Vienna international workshop on
functional electrostimulation, vol. 8. 2004. p. 1914.
[7] Lawrence M, Kirstein T, Keller T. Electrical stimulation of nger exors using
virtual electrodes. In: Vienna international workshop on functional electrostimulation, vol. 8. 2004.
[8] Keller T, Lawrence M, Kuhn A, Morari M. New multi-channel transcutaneous
electrical stimulation technology for rehabilitation. In: Annual international
conference of the IEEE engineering in medicine and biology society. 2006. p.
1847 (IEEE Cat. No. 06CH37748).
[9] Lawrence M, Pitschen G, Keller T, Kuhn A, Morari M. Finger and wrist torque
measurement system for the evaluation of grasp performance with neuroprosthesis. Artif Organs 2008;32(8):6348.
[10] Keller T, Kuhn A, Lawrence M, Morari M. Textile neuroprosthesis garment for
functional electrical stimulation. In: Vienna international workshop on functional electrostimulation, vol. 9. 2007.
[11] Kuhn A, Keller T, Lawrence M, Morari M. A model for transcutaneous
current stimulation: simulations and experiments. Med Biol Eng Comput
2009;47(3):27989.
[12] Kuhn A, Keller T. The inuence of capacitive properties on nerve activation in
transcutaneous electrical stimulation. In: International Symposium on Computer Methods in Biomechanics and Biomedical Engineering, vol. 7. 2006.
[13] Plonsey R, Heppner DB. Considerations of quasi-stationarity in electrophysiological systems. Bull Math Biophys 1967;29(4):65764.
[14] Kuhn A, Keller T, Prenaj B, Morari M. The relevance of non-linear skin properties
for a transcutaneous electrical stimulation model. In: International Functional
Electrical Stimulation Society Conference, vol. 11. 2006.
[15] Gabriel S, Lau RW, Gabriel C. The dielectric properties of biological tissues:
III. Parametric models for the dielectric spectrum of tissues. Phys Med Biol
1996;41(11):227193.
[16] Reilly JP, Antoni H, Chilbert MA, Sweeney JD. Applied bioelectricity from electrical stimulation to electropathology. New York: Springer; 1998.
[17] Rijkhoff NJ, Holsheimer J, Koldewijn EL, Struijk JJ, van Kerrebroeck PE, Debruyne
FM, Wijkstra H. Selective stimulation of sacral nerve roots for bladder control:
a study by computer modeling. IEEE Trans Biomed Eng 1994;41(5):41324.

A. Kuhn et al. / Medical Engineering & Physics 31 (2009) 945951

[18] Prodanov D, Feirabend HK. Morphometric analysis of the ber populations of
the rat sciatic nerve, its spinal roots, and its major branches. J Comp Neurol
2007;503(1):85100.
[19] Lertmanorat Z, Gustafson KJ, Durand DM. Electrode array for reversing the
recruitment order of peripheral nerve stimulation: experimental studies. Ann
Biomed Eng 2006;34(1):15260.
[20] McIntyre CC, Richardson AG, Grill WM. Modeling the excitability of mammalian
nerve bers: inuence of afterpotentials on the recovery cycle. J Neurophysiol
2002;87(2):9951006.
[21] Warman EN, Grill WM, Durand D. Modeling the effects of electric elds on
nerve bers: determination of excitation thresholds. IEEE Trans Biomed Eng
1992;39(12):124454.

951

[22] Moftt MA, McIntyre CC, Grill WM. Prediction of myelinated nerve ber
stimulation thresholds: limitations of linear models. IEEE Trans Biomed Eng
2004;51(2):22936.
[23] Standring S. Grays Anatomy. 39th ed; 2005.
[24] Lyons GM, Leane GE, Clarke-Moloney M, OBrien JV, Grace PA. An investigation of
the effect of electrode size and electrode location on comfort during stimulation
of the gastrocnemius muscle. Med Eng Phys 2004;26(10):8738.
[25] Keller T, Hackl B, Lawrence M, Kuhn A. Identication and control of hand
grasp using multi-channel transcutaneous electrical stimulation. In: International functional electrical stimulation society conference, vol. 11. 2006.
p. 2931.