IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 41, NO.

7 , JULY 1994

68 I

A Nonlinear Finite Element Model of the Electrode-Electrolyte-Skin System

Dorin Panescu, Member-, IEEE, John G. Webster, Fellow, IEEE, and Robert A, Stratbucker, Member, IEEE
Abstract-This study presents a two-dimensional finite element model of the electrode-electrolyte-skin system which takes into account the nonlinear behavior of the skin with respect to the amplitude of the voltage. The nonlinear modeling approach has practical value for studies related to transcutaneous stimulation (e.g. maximizing the dynamic range of sensory substitution systems, optimization of TENS, optimization of transcutaneous cardiac pacing, etc.). The model has three main regions: 1) the electrolyte; 2) the skin; and 3) the body. The model consists of 364 nodes, 690 elements and was generated on a MacIntosh I1 using a version of FEHT (Finite Element for Heat Transfer) adapted for electromagnetics. The electrodes are equipotential lines and the electrolyte is modeled as a pure resistive region with constant conductivity. Although the electrode-electrolyte interface can introduce nonlinearities, we did not take them into account because the skin displays a much higher impedance. The skin is modeled as a nonlinear material with the conductivity dependent on the applied voltage. To account for the mosaic structure of the skin, we used ten different nonlinear subregions of five different values of breakdown voltage. The region designated body models the effects of the resistance associated with the dermis and the tissues underneath the skin, and has a constant high conductivity. We studied the effects of two different electrolytes on the comfort of stimulation and found that there was less potential pain delivered when high-resistivity electrolytes were used. This was due to the larger nonuniformities in the current density distribution which appeared for low-resistivity electrolytes. Moreover, increasing the skin temperature made the current density even more nonuniformly distributed for lowresistivity electrolytes. Experiments performed on the skin of the left arm, using 1-cm Ag-AgCI electrodes, showed that the skin broke down at spots of lowest breakdown voltage. This is consistent with reports of previous experimental studies and has practical value for the design of optimal electrodes. type of model has limited practical value since the current density underneath the electrode is not uniformly distributed [l]. Therefore, a lumped electrical model would not be able to predict, for example, under which part of the electrode the skin breakdown is more likely to occur. The lumped models, as well as one-dimensional electric networks used to model the skin, also have the disadvantage that they can not quantify the current density vector (i.e., its magnitude and direction). Moreover, a one-dimensional electric network has to include many nonlinear devices to yield resolution comparable with a nonlinear finite element model. The circuit simulators employed to analyze this network (e.g. SPICE) might have convergence problems for a large number of nonlinear devices. Also, the display of the current density vectors during the postprocessing phase might be difficult to achieve. Although approximate analytical solutions to this the problem have been reported [ 2 ] , skin is usually considered to have a constant conductivity and its mosaic structure is not taken into account. This study proposes a new approach of assessing the current flow through the electrode-electrolyte-skin system based on nonlinear two-dimensional finite element analysis. The electrolyte and the tissues underneath the skin can be modeled with good approximation as linear regions [3]. Moreover, until it breaks down, the skin mainly determines the resistance of the system [3], [4]. Therefore, we modeled only the skin as a nonlinear material, assigned constant conductivities to the other elements of the system, and did not take into account the nonlinearities of the electrode-electrolyte interface. The model also accounted for the effects of the mosaic electrical structure of the skin. The current flow through the skin has a nonlinear dependence on the voltage across it [ 5 ] , [6]. Hereafter, we consider the voltage-current characteristic (VCC) of the skin to be: 1) quasilinear, 2) nonlinear and symmetric, or 3) nonlinear and asymmetric if there is a corresponding 1) quasilinear, 2) nonlinear and symmetric, or 3) nonlinear and asymmetric, relationship between the biphasic current flowing through the skin and the symmetric biphasic voltage across it. For low amplitude voltages or for high frequencies, the VCC of the skin is quasilinear. As the amplitude of the stimulation increases or the frequency decreases, the VCC of the skin becomes nonlinear and symmetric. For large voltages or for very low frequencies, the VCC becomes asymmetric. The sudden decrease of the skin impedance, causing current runaway, is known as electrical breakdown of the skin. As shown in [6], the onset of the breakdown

I. INTRODUCTION

HE DESIGN of optimal electrodes for transcutaneous electrical stimulation usually requires the investigation of current flow through the skin. Although it is possible to measure the current density undemeath the electrodes directly, this requires complex instrumentation. The direct measurement is also time consuming because it requires mounting fine probes undemeath the electrode or skin. Usually lumped models of the electrode-electrolyte-skin system are used for an initial assessment of this current flow. However, this
Manuscript received May 21, 1992; revised November 26, 1992, and March
1. 1994. Supported by a grant from Marquette Electronics, Incorporated.

D. Panescu is with E P Technologies, Incorporated, Sunnyvale CA 94086 USA. J. G. Webster is with the Department of Electrical Engineering, University of Wisconsin, Madison, WI 53706 USA. R. A. Stratbucker is with the Radiation Health Center of the State of Nebraska, Omaha, NE 68 152 USA. IEEE Log Number 9402026.

0018-9294/94$04.00 0 1994 IEEE

682

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 41, NO. 7, JULY 1994

occurs at the transition of the skin VCC from the nonlinear and symmetric into nonlinear and asymmetric. After this transition, the subject experiences a stinging sensation of pain. We used cyclic voltammograms, similar in shape to Lissajous figures, to describe the VCCs of the skin because they provide information about the imaginary part of the skin impedance, as well as its real part. A quasilinear VCC implies a corresponding cyclic voltammogram very close to an ellipse. For a nonlinear and symmetric VCC, the cyclic voltammogram is a distorted ellipse with identical distortions for the positive and negative phases. A system with a nonlinear and asymmetric VCC yields a cyclic voltammogram that is a distorted ellipse with the positive phase distortion different from the negative phase distortion. The skin has a mosaic electrical structure mainly because the stratum comeum is crossed by sweat ducts, which are the main current pathways and nonuniformly distributed on the surface of the skin [l], [7]-[ll]. To demonstrate the mosaic electrical structure of the skin, [9] reported that the impedance of the skin had very different values when it was measured using a suction microelectrode at spots that were only few hundreds of micrometers apart. Webster [ l ] showed that, on a microscopic scale, the current passes through many small pores. When the applied voltage became large enough, one or more of these pores broke down generating the current runaway. Grimnes [ 101 presented evidence that the main current pathways through the skin were the sweat ducts. Based on this information, hereafter, the sweat ducts will be considered the regions of lowest breakdown voltage. We studied the effects of two different electrolytes on comfort of stimulation and found that it was potentially less painful to stimulate when high-resistivity electrolytes were used. When the temperature of the skin increased, for the low-resistivity electrolyte case, the current became even more nonuniformly distributed, potentially causing more pain, but for the high-resistivity electrolyte its distribution was not significantly affected. The design of electrodes for transcutaneous stimulation should choose the optimal combination of electrode size and electrolyte resistivity which yields the lowest ratio between the maximal and minimal current densities through the skin, and, therefore, the most uniform current distribution.

finite element computations is solved based on the Cholesky factorization method. We considered the biological tissues as continuous materials, their conductivity being averaged over a volume containing many cells. Each of the regions within the finite element model (FEM) was considered to be time-invariant, isotropic and free of electric charges. We were interested in the values of the current density, given by

and because J = a E and E = -VV

v . (aVV) = 0

(2)

11. THE NONLINEAR TWO-DIMENSIONAL FINITEELEMENT MODEL To perform our finite element computations we employed FEHT (Finite Element for Heat Transfer) [ 121, a program developed for solving heat transfer and electromagnetic problems on Apple MacIntosh series computers configured with a minimum of 1 MB of main memory. To build a finite element model, the user has to manually generate a coarse mesh, the size of which can be automatically reduced within the regions of interest. FEHT uses triangular elements and first-order basis functions. Before starting to calculate the solution of any forward problem, FEHT checks if the mesh is free of errors (i.e., intersecting edges and nodes closer than a prespecified tolerance). The system of linear equations involved in the

where J is the current density in [A/m2], E is the electric field intensity in [V/m], V is the electric potential in [VI and a is the material conductivity in [S/m]. For constant a, (2) becomes a Laplace equation. To solve (2) (i.e., to find the electric potential), one of the following types of boundary conditions can be used: 1) V is specified all the way around the model outline; 2) the normal current density is specified all the way around the model outline; 3) V is specified only on some part of the model outline and the normal current density on the rest of it. The function V which satisfies (2) and one of the conditions l), 2) or 3) is uniquely determined. The boundary value problems of type 1) are known as Dirichlet problems, the ones of type 2) as Neumann problems, and the ones of type 3) as mixed problems. We solved mixed problems by specifying the voltage over the region corresponding to the electrodes and set the normal current density to zero on the outline between the electrodes. The region corresponding to the skin had a conductivity dependent on the potential, thus, (2) was nonlinear. EEHT employed an iterative algorithm to solve this nonlinear equation. Initially, (2) was solved for the skin conductivity equal to the value corresponding to 0 V across it. Then, based on these computed values of the potential, the skin conductivity was updated. The iterations continued until the Euclidean norm, of the difference between subsequent solutions of the potential inside the model, was less than a chosen tolerance. The current density was further computed applying J = -aVV inside each element of the model. The last value in the series of updated skin conductivities was used to find J for the region modeling the skin. We checked the accuracy of FEHT for solving nonlinear problems by comparison against the analytic solution for a simple model. For a homogeneous rectangular model, consisting of a single type of nonlinear material, for which the current-voltage equation is

(3)
where I is the current flowing through the model expressed in microamperes, IO is a constant chosen to be 1.5 pA and V is the voltage across the model expressed in volts, the

PANESCU er al.: A NONLINEAR FINITE ELEMENT MODEL OF THE ELECTRODE-ELECTROLYTE-SKIN SYSTEM

683

TABLE I THE MAXIMAL ERRORS VERSUS NUMBER THE OF NODES THE SIMPLE IN NONLINEAR MODEL Number of nodes 33 Maximal error 3%

113 0.5%

417 0.4%

conductivity is equal to 1 exp g ( V ) = Io--

(F)

(4)

where 1 and S denote the depth and the area of our rectangle. The voltage, V, was swept from 0 to 50 V. Table I lists the maximal errors, between the theoretical value of the current and the value returned by FEHT, as a function of the number of nodes in the model. We concluded that for more than 200 nodes in the nonlinear region the accuracy of FEHT was suitable for our purposes. Fig. 1 shows the lumped model for the skin proposed by [6]. The current flow through the skin is mainly affected by iontophoretic rather than electroosmotic mechanisms. Therefore, different ion species account for the outflow than for the inflow through the skin. D1 and D2 are nonlinear devices to simulate both the nonlinear and asymmetric behavior of the skin. They are described by

Fig. 1. The electrical model of the skin. RCOrllellr., corresponds to the resistance through the stratum corneum and it has a high initial value. When the temperature of the skin increases its value decreases. D1 and Dz are nonlinear devices to account for the nonlinearities and asymmetries in the electrical characteristic of the skin. C,k is the equivalent capacitance of the stratum corneum and increases with temperature of the skin. Rbodl corresponds to the resistance of the dermis and has a low value. THE PARAMETERS FOR Parameter Average of 1 0 IO k1 k21 (for the positive phase) li22 (for the negative phase) k3 TABLE I1 ELECTRICAL MODEL IN Value
FIG. 1

THE

sv
1.5 p A

2 0.38 0.17 0.003

where Io is the value of I for large negative V when the skin considered to be temperature, T , is at its normal value, Enit., 34C [13], and f l i ( T ) ,f 2 ( T ) are functions of skin temperature [61

(7)

where V is the value of f 2 ( T ) when the skin is at its o normal temperature. The regions of skin with different breakdown voltage were assigned different VO.The parameters V~:Io,kl,kzl(kl D 1 ) , k 2 2 ( k 2for D z ) , and ks are given for is in Table 11. RcOrneum the resistance corresponding to the leakage through the stratum corneum and has a very high value. Csk is the equivalent capacitance of the stratum corneum. R b o d y is the resistance associated with the dermis and it has a value of about 1 kR [3]. Except for R b o d v , all the other elements in this model have their values dependent on the skin temperature. As [6] shows, when the VCC of the skin is asymmetric its impedance is mostly determined by devices D1 and D2 in series with Rbody. Therefore, in the finite element model, the regions of skin are formed only by the antiparallel connected diode-like devices D1 and D2. Thus, the current flowing through the skin in our model is given by

When the skin temperature increases, for a constant voltage V , the current through the skin I , also increases. Therefore, the resistance of the skin decreases as its temperature increases. Fig. 2 shows the two-dimensional finite element model (FEM) we generated for this study. It consists of three main regions: 1) the electrolyte (gel); 2) the skin; and 3) the body. The stimulating electrode was placed on top of the electrolyte, whereas the reference electrode was placed at the bottom end of the body region. The model comprised 364 nodes and 690 elements and its dimensions were roughly scaled up by 10, with respect to the experimental setup, where we used a 1-cm2 stimulating electrode. Out of these nodes, 218 were in the regions corresponding to skin. The meshes for the gel and body had a higher density of nodes at the boundary with the skin to ensure a smooth transition between the regions of different conductivities. The electrodes were equipotential. The conductivities of the regions corresponding to the electrolyte and the body were constant. The conductivity of the skin depended nonlinearly on the applied voltage, based on (8). Table I11 shows the conductivities of the regions in our FEM. We studied the effects of two types of electrolyte (gel): a) the low-conductivity electrolyte yields a 1.5-kR equivalent lumped resistance of the gel layer; and b) the high-conductivity electrode yields resistance of 80 R. Their effects upon the current density through the system have been analyzed at two temperatures of the skin: T = 34C and T = 40C. As Fig. 2 shows, to simulate the mosaic structure of the skin, we divided it into ten regions of conductivity, each containing an average of 20 nodes of the finite element mesh. These regions had the following values of VO (at the normal temperature of the skin, T = T,,,it,): avg had VO= 5 V (breakdown voltage of 15 V); 1 had V = 4 V (breakdown voltage of , 12 V); 2 had VO = 4.5 V (breakdown voltage of 13.5 V); 3 had V, = 5.5 V (breakdown voltage of 16.5 V); 4 had V, = 6 V (breakdown voltage of 18 V). For the skin, Table 1 1 gives the conductivities at T = Tin;,,(34C). The values 1

68-1

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 41. NO. 7 , JULY 1994

F I ~ . 2. The two-dimensional finite element model of the electrode-electrolyre-skin system. To account for the mosaic structure of the skin. we divided it into ten regions of five different values for the hreakdown voltage.

THECOUULICTIVITIES

OF THE

TABLE Ill REGIONS THE IN

FINITE

ELEMENT MODEL

Region Gcl (high-re\istivity) Gel (low-re5istivity) Skin "avg" at 34OC Skin "1" at 34OC Skin "2" at 34OC Skin "3" ar 34C Skin "4" at 34C Body

Conductivity 0.003 S/m 0.06 S/m 0.7he-h/V*(exp(V/S V ) + exp(-V/S V)) S/m 0.76e-6/V*(exp(V/4 V) + exp(-V/4 V)) S/m 0.76e-6/V*(exp(V/4.5 V) + exp(-V/4.5 V)) S/m 0.76e-6/V*(exp(V/5.5 V ) + exp(-V/5.5 V)) S/m 0.76e-6/V*(exp(V/6 V) + exp(-V/6 V)) S/m 0.7 S/m

(b)

corresponding to T = 40C can be computed by making use of ( 6 ) ,(7) and (S), and they are higher than those in Table 111. We chose two regions of lowest breakdown voltage because, in the experiment presented in Section 111, the skin broke down at five spots underneath the 1 -cm2 electrode. The results reported by [SI and [lo] motivated us to consider these spots as the lowest breakdown voltage regions of skin. Hence, these yield about two regions of lowest breakdown voltage per 1-cm of linear dimension. The "body" was modeled to have a very low equivalent lumped resistance [ 3 ] .We assessed the effects of the proximity of the vertical Neumann type boundaries by performing simulations with increased distance between them and the edge of the gel. The results were not significantly different from those corresponding to the configuration in Fig. 2.
111. SIMULATION A N D EXPERIMENTS

Fig. 3. (a) The current flow through the model at 10 V for the high-resistivity electrolyte, when the skin temperature was 34OC. The ratio between the maximal and minimal current densities in the skin undemeath the electrode was 2.7: (b) the equipotential contours. The entire voltage was developed across the thickness of the skin because of its very high resistance. The regions " I " are hatched.

We performed simulations at two temperatures of the skin, 34C and 40C, for two types of electrolyte (gel). The voltage across the system varied from 5 V to above 50 V, in order to cover all of the quasilinear, nonlinear and symmetric, and asymmetric ranges.
A . Siniidatioris Usirig the High-Resistility Elrctrnlyte

At 5 V applied across the system the ratio R, between the highest and lowest current densities in the skin undemeath the electrode, was 1.79, when the skin temperature was 34"C, and 1.65, at 40C. The current was rather uniformly distributed, and the whole voltage was developed across the skin.

At 10 V applied across the system, R was 2.7 when the skin temperature was 34"C, and 2.2, at 40C. The current was, again, rather uniformly distributed and the whole voltage was developed across the skin. Fig. 3(a) shows the current density distribution in the model for the skin temperature equal to 34C. Fig. 3(b) shows the equipotential contours at 10 V for the skin temperature equal to 34C. Note that all the 10 V are developed across the skin as we would expect (i.e., because of the very large resistance of the skin). At 22 V applied across the system, R was 6.23, when the skin temperature was 34"C, and 6.06, at 40C. The current was no longer uniformly distributed; a larger amount was flowing through regions "1 ," which were the lowest breakdown voltage regions. However, most of the applied voltage was still developed across the skin. Above 22 V, the higher skin conductivity at 40C made the current slightly more nonuniformly distributed than at 34C. At 50 V applied across the system, the ratio R was 14.87, when the skin temperature was 34"C, and 15.7, at 40C. The current was funneled through regions "1" where the skin current density reached its maximum of 0.56 mA/cm2, at 34"C, and 0.72 mA/cm2 at, 40C. Fig. 4(a) shows the current density distribution and Fig. 4(b) shows the equipotential contours for this case, both at 34C. The length of the shafts of the arrows varies linearly with the current density, from its minimum, shown in Fig. 3(a), to the values in Fig. 4(a). Note that the major part of the applied voltage was developed across the electrolyte. Across the regions "1" only 4 V, at 3 4 C and

PANESCU

et

al.: A NONLINEAR FINITE ELEMENT MODEL OF THE ELECTRODE-ELECTROLYTESKIN SYSTEM

voltage was still developed across the skin. At 40C and above 20 V, the higher skin conductivity made the current more nonuniformly distributed than at 34OC. Also, the changes due to the temperature increase were greater than for the high-resistivity electrolyte. When we injected above the same amount of current into the system as for 50 V applied across the system with the high-resistivity electrolyte, the ratio R was 33.7, at 34C. and 40.1 V, at 40C. The voltage drop across regions 1 was 41 V. at 3 4 C and 6.5 V, at 40C. The voltage across the regions in the middle of the model was 47 V, at 34C and 1 1 V, at 40C. Fig. 5(a) shows the current density distribution and Fig. 5(b) shows the equipotential contours for the skin temperature equal to 34C. More current is funneled through the regions of lowest breakdown voltage than for the case when the highresistivity electrolyte was used. The current density reached a maximum of 5.9 mA/cm2. at 34C. and 11.8 mA/cm2, at 40C, inside regions I . There is no relation between the length of the shafts of the arrows in Fig. 4(a) and in Fig. S(a). If a proportional representation were used, the shafts in Fig. 5(a) would have been about ten times longer. The maps of the current density and equipotential contours looked similar at 40C, therefore we did not include figures.
(b) Fig. 4. (a) The current flow through the model at 50 V for the high-resistivity electrolyte, when the skin temperature was 34OC. The current density distribution was much more nonuniform and the current flows mostly through the two regions of lowest voltage breakdown. The ratio X was 14.87 and the maximal current density was 0.56 mA/cm2; (b) the equipotential contours. The voltage developed across the thickness of the two regions of lowest voltage breakdown was about 4 V. The voltage developed across the thickness of the skin in the middle of the model was about 20 V. Anomalies in arrow directions may be caused by current flowing around regions having high breakdown voltages. Shaft length representing minimal current density is short. Length increases linearly with current density. The regions I are hatched.

C . Espet-inient to Check the Hypothesis that the Skin Initially Breaks DoMn at the Regions of Lovt9est Breakdovtw Voltuge
Both of the simulations in Sections 1II.A and B showed that the skin initially broke down at the regions of lowest breakdown voltage. This suggests that for a test on the skin, we should observe the breakdown initially occurring only at several spots underneath the stimulating electrode. To test this assertion, we built two I-cm Ag-AgC1 (because Ag-AgC1 approaches the characteristics of perfectly nonpolarizable electrodes (41) square electrodes from a larger tab ECG electrode. We tested the skin on the left forearm. The reference electrode, a pregelled 36-cm2 Ag-AgC1 electrode, was applied in the region of the left biceps muscle. To have reproducible experiments, to keep the electrode in place and to distribute the pressure uniformly, we used a small rubber band between the shaft of a force gage and the electrode, and applied a constant force of 0.5 N. First, we applied a sine wave of IO-V amplitude and 20-Hz frequency. Then, after the skin recovered for about 15 min, we increased the amplitude to 20 V. The instrumentation acquired the VCC of each individual electrode and then that of their parallel combination. The VCCs are displayed using cyclic voltammograms. To prevent changes in the characteristic of the skin due to its recovery, the whole test lasted about 45 s. Fig. 6(a) shows that at 10-V amplitude. the breakdown did not occur and by connecting the electrodes in parallel. thus increasing the area twofold. the skin admittance increased approximately twofold. Fig. 6(b) shows that at 20-V amplitude the skin broke down under electrode # 2 and increasing the area twofold, by connecting the electrodes in parallel. the admittance of the system did not increase twofhld and was about equal to the admittance corresponding to the electrode under which the skin broke down. At the onset of the skin

0.8 V, at 40C, were developed, while across regions in the middle of the model the voltage was about 20 V, at 34C, and 1.3 V, at 40C. The difference was a consequence of the fact that the skin first broke down at the regions 1, which had the lowest voltage breakdown. At 40C, the skin resistance was significantly reduced, therefore, the voltage across it decreased as well. The display of the current density and equipotential contours looked similar for 40C, therefore we did not include figures. B. Simulations Using the Low-Resistiiity Electmlyre At 5 V applied across the system, the ratio R was 1.78, when the skin temperature was 34C, and 1.7, at 40C. The current was rather uniformly distributed and the whole voltage was developed across the skin. At 10 V applied across the system, R was 2.75, when the skin temperature was 34C, and 2.25, at 40C. The current was, again, rather uniformly distributed, and the whole voltage was developed across the skin. Because the current distribution and the equipotential contours were very similar to those for the high-resistivity electrolyte we did not include figures. At 28 V applied across the system, R was 7.67, when the skin temperature was 34C, and 7.8, at 40C. The current was no longer uniformly distributed, a larger amount was flowing through regions 1 . However, most of the applied

686

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 41, NO. 7, JULY 1994

-20

-15

-10

-5 0 Voltage [VI

10

15

(a)
700
500

300

(b) (a) The current flow through the model for the low-resistivity Fig. 5. electrolyte, at 34OC, when the injected current was about the same as that in Fig. 4 (E 60 mA). The ratio R was 34.2 and more current was funneled through the regions of lowest breakdown voltage. The highest current density was 5.9 mA/cm, about ten times larger than for the high-resistivity electrolyte; (b) the equipotential contours. The voltage developed across the thickness of the two regions of lowest voltage breakdown was about 41 V. The voltage developed across the thickness of the skin in the middle of the model was about 47 V. Anomalies in arrow directions maybe caused by current flowing around regions having high breakdown voltages. Shaft length representing minimal current density is short. Length increases linearly with current density. The regions 1 are hatched.

-100

-300
-500

c]
4
-20 -15 -10

-700

-25

-5

0 5 Voltage [VI

10

15

20

25

(b) Fig. 6. (a) The VCC of the parallel combination of two 1-cm2 electrodes and of each of the electrodes before the breakdown of the skin occurred; (b) the VCCs for the same set up after the skin broke down undemeath electrode #2. Once the skin was broken down, the characteristic of the system was determined mainly by the region where the breakdown occurred.

breakdown, the subject experienced a stinging pain sensation. Looking at the skin under electrode #2, immediately after the experiment was over, we noticed five black tiny spots with a reddish circumference. The skin was permitted to recover for about 2 h. To test that these spots were the lowest breakdown voltage sites of the tested location, we set the amplitude to 20 V and, as presented above, observed the breakdown initially occurring underneath electrode #2 (i.e., its cyclic voltammogram displayed the current runaway as shown in Fig. 6(b)). The voltage amplitude was then increased to 22 V and this generated the current runaway underneath electrode # I as well. Whereas at the onset of the skin breakdown undemeath electrode #2 the pain sensation could be tolerated, when the skin undemeath electrode #1 broke down the pain was too intense to be tolerated, therefore, the test was immediately interrupted.
IV. DISCUSSION

The experiment presented herein confirmed the hypothesis that, initially, the breakdown of the skin occurred at the regions of lowest breakdown voltage, which are believed to be the sweat ducts. Also, it showed that the conductance of the

electrode-skin system was approximately proportional to the system area only until the breakdown occurred. After that, the relationship was dictated by that part of the system which broke down. We found five spots of lowest breakdown voltage under electrode #2, which can be approximated with two such spots per 1 cm of linear dimension. For this reason, we included two regions of lowest breakdown voltage in our finite element model. Fig. 7 plots the dependence of the ratio R on the current injected into the system for each of the electrolytes at both the normal temperature of the skin, 34C, and at 40C. Although some electrolytes have a negative temperature coefficient of resistivity, because its value is of only few percentPC, its effects were not accounted for. Note that the low-resistivity electrolyte yielded much larger values for the ratio R, hence, more nonuniform current distribution than the high-resistivity electrolyte. Also, when the temperature increased the ratio R, hence the nonuniformity of the current distribution, increased more for the low-resistivity electrolyte than for the high-resistivity one. Also in Fig. 5(b), note that the voltage developed across the skin was larger than that presented in Fig. 4(b) for the high-resistivity case,

PANESCU et al.: A NONLINEAR FINITE ELEMENT MODEL OF THE ELECTRODE-ELECTROLYTE-SKIN SYSTEM

687

40

30

20

a finite set of practical electrolyte resistivities for the specific application. 4) To perform finite element simulations for each combination of electrode size and electrolyte resistivity and choose the one that yields the minimal ratio R (maximal over minimal current density in the skin). Examples of specific applications that could benefit from the design of optimal electrodes are: the optimization of the dynamic range of stimuli for a sensory substitution system, transcutaneous electrical nerve stimulation, and transcutaneous cardiac pacing.
0 0

10

h h resistivitygel (WC) u high resistivky gel ( W C ) b w resistivity. (WC) . gel . . b w resiotwity gel ( W C )

REFERENCES
J. G. Webster, Minimizing cutaneous pain during electrical stimulation, in Proc. 9th Annu. Con$ IEEE Eng. Med. Biol. Soc., vol. 9, pp. 986-987, 1987. C. M. Reddy and J. G. Webster, Uniform current density electrodes for transcutaneous electrical nerve stimulation, in Proc. IEEE Frontiers of Engineering and Computing in Health Care Con$, pp. 187-190, 1984. E. Freiberger, The electrical resistance of the human body to commercial direct and alternating currents. Berlin: Verlag van Julius Springer, 1934. M. R. Neuman, Biopotential electrodes, in Medical Instrumentation: Application and Design, J. G. Webster, Ed., 2nd ed. Boston: Houghton Mifflin, 1992. T. Yamamoto and Y. Yamamoto, Nonlinear electrical properties of skin in the low frequency range, Med. Biol. Eng. Comput., vol. 19, pp. 302-310, 1981. D. Panescu, J. G. Webster and R. A. Stratbucker, A nonlinear electricalthermal model of the skin, IEEE Trans. Biomed. Eng., vol. 41, no. 7, pp. 672-680, July 1994. D. P. Burbank and J. G. Webster, Reducing skin potential motion artefact by skin abrasion, Med. B i d . Eng. Comput., vol. 16, pp. 31-38, 1978. A. van Boxtel, Skin resistance during square-wave electrical pulses of 1 to 10 mA, Med. B i d . Eng. Comput., vol. 15, pp. 679-687, 1977. D. Panescu, K. P. Cohen, J. G. Webster and R. A. Stratbucker, The mosaic electrical structure of the skin, IEEE Trans. Biomed. Eng., vol. 15, pp. 4 3 4 4 3 9 , 1993. S . Grimnes, Pathways of ionic flow through human skin in vivo, Acta Derm. Venereol., vol. 64, pp. 93-98, 1984. T. Yamamoto, Y. Yamamoto and A. Yoshida, Formative mechanism of current concentration and breakdown phenomena dependent on direct flow through the skin by a dry electrode, IEEE Trans. Biomed. Eng., vol. BME-33, pp. 396-404, 1986. S. A. Klein, W. A. Beckman and G . E. Myers, FEHT a finite element analysis program for heat transfer and electromagnetics, F-Chart Software, Middleton WI, 1989. V. C. Rideout, Mathematical and Computer Modeling of Physiological Systems, Englewood Cliffs NJ: Prentice-Hall, 1991. J. E. Monzon, Noninvasive cardiac pacing electrodes, M.S. thesis, Univ. of Wisconsin, Madison, WI, 1986. P. M. Zoll, R. M. Zoll and A. H. Belgrad, External noninvasive electric stimulation of the heart, Crit. Care Med., vol. 9, pp. 393, 1981. J. P. Reilly, Electrical Stimulation and Electropathology. Cambridge: Cambridge Univ. Press, 1992. S. Grimnes, Dielectric breakdown of human skin in vivo, Med. Biol. Eng. Comput., vol. 21, pp. 379-381, 1983. G. B. Kasting and L. A. Bowman, DC electrical properties of frozen, excised human skin, Pharm. Res., vol. 7, pp. 134-143, 1990.

5000

10000

15000

Stimulating current, 1.5e-2 mA

Fig. 7. The ratio of Jmax/Jmin as a function of the stimulating current for the two types of electrolytes we used, at the normal temperature of the skin, 34OC, and at 4OOC. The low-resistivity electrolyte yielded much larger ratio R and the current was even more nonuniformly distributed when the skin temperature increased.

even though the current injected into the system was about the same (60 mA). The local current density was also higher. This means that more heat is dissipated in the skin when lowresistivity electrolytes are used. As [6] and (6), (7) and (8) show, the increase of the skin temperature favors the onset of the breakdown by further reducing its electric resistance. Therefore, the low-resistivity electrolytes contribute to a faster onset of the current runaway and, thus, of the breakdown, by two paths: 1) the larger nonuniformity of the current flow through the skin; and 2 ) larger heat dissipation in the skin, which causes the increase of the local skin temperature, which, therefore, increases the ratio R, hence, makes the current even more nonuniformly distributed. The onset of the breakdown generates an uncomfortable sensation or even pain. Thus, using low-resistivity electrolytes can be more painful for the patient who undergoes the transcutaneous stimulation. This assertion is consistent with the results presented by Monzon [ 141 who found experimentally that the onset of pain occurs faster when low-resistivity electrolytes are applied, and also with the results presented by Zoll et al. [15]. Knowing how the electrolyte may affect the current distribution pattem is useful for the design of optimal electrodes. Webster [ l ] shows that it is less painful to transcutaneously stimulate using an electrode with 800-R face-to-face impedance than with 8-0 face-to-face impedance. The finite element method presented herein could be used for optimal electrode design as follows: 1) To specify a finite set of practical electrode areas for the specific application to be optimized. As shown in Section III.C, it is important to check the effects of the electrode size because its conductance is no longer proportional to the area once the VCC of the system becomes nonlinear and asymmetric. 2 ) To account for the mosaic electrical structure of the skin by randomly distributing as many regions of lowest breakdown voltage as predicted by the average density of sweat ducts for the site of the skin involved in the specific application [16]. 3) To specify

Dorin Panescu (S92-M94), for photograph and biography, see p. 679 of

this issue.

John G. Webster (M59-SM69-F86), for photograph and biography, see

p. 680 of this issue.

Robert A. Stratbucker (S54-M58), for biography, see p. 680 of this issue.