You are on page 1of 4

Proceedings of the 22"dAnnual EMBS International Conference, July 23-28, ;!OOO, Chicago IL.

THE USE OF ELECTRICAL IMPEDANCE TOMOGRAPHY WITH THE INVERSE PROBLEM OF EEG AND M[EG

s.GonGalves***'and J.C. de Munck*


'MEG Center KNAWNniversity Hospital Vrije Universiteit, Amsterdam, The Netherlands
88

Institute of Biophysics and Biomedical EngineeringFaculty of Sciences, Lisbon.,Portugal

1. Introduction
The Inverse Problem ( I F ' ) of EEG [I] and MEG [2] aims to estimate the sources inside the brain based on measurements of the electrical potential on the scalp surface (in the case of the EEG) or on measurements of the magnetic field outside the head (in the case of the MEG). In the solution of the IP it is assumed that the head is divided into a number of compartments, and that the conductivities of these compartments are known. The head can be modelled as a set of concentric spheres [3], each sphere representing each of the head compartments, or by using a realistic model, which consists of a set of nested triangulated surfaces. In both types of models, the conductivities of the different head compartments are included as known constants. In the EEG case it appears that the solution of the IP [4]-[7] is highly dependent on the values of these constants. Therefore large systematic errors in the localisation of the sources may arise due to errors in the values of the electrical conductivities. In practice, the values of the electrical conductivities are taken from literature and they result from very few conductivity measurements which, most of the times, were performed on isolated samples of tissue. These values vary over a wide range and there might be a factor of 7 between the maximum and minimum conductivity values reported for a certaiin tissue [8],[9]. This wide variation of the conductivity values is related to the variation of tissue properties between individuals, to the fact that the tissues are neither homogeneous nor isotropic and also to the variation in measuring techniques. Electrical Impedance Tomography (EIT) rlO] is a technique used to determine the distribution of electrical conductivities inside an object from the measurement of the electrical potential distribution at the surface, which results from the injection of currents into the object. In practice, the current is injected into a pair of electrodes and measured on the remaining. In the case of EEG IP no detailed distribution of the conductivities is needed. On the contrary, the knowledge of the equivalent conductivities of the head compartments is sufficient to srolve the source localization problem. In this paper the principles of EIT are applied to determine these equivalent conductivities and it is shown that their use in solving the EEG IP decrease the systematic errors of the solution.

2. Materials and Methods


In this study, the model represented in figure 1 was used to model thle head. In this situation the solution of the fclrward problem (i.e. given a grid of electrodes placed at the outer sphere, the conductivies values and the injected electrical current, determine the potential distribution at the outer surface) was obtained by using a spherical harmonics expansion of the potential. Electrodes of finite dimensions were considered and the distribution of the current injected at a given electrode was assumed to be given by:

In (1) 0 is defined in figure 2.a, 0, is the angular dimension of the injecting electrode as defined in figure 2.b and k is a constant which is adjusted such that:

jjJ@)R
dF

sin(8)tle d q = 0

(2)

0-7803-6465-1/00/$10.00 02000 IEEE

2346

Proceedings of the 22"dAnnual EMBS International Conference, July 23-28, 2000, Chicago IL.

Figure 1. Schematic representation of a 3 layer sphere of radius RI, R2, R3 and conductivities ol, 02, 0 3 .
Solving the problem in the situation considered in figure 2.b by imposing the appropriate boundary conditions one arrives at a linear relation between the injected current and the potential generated at the surface:

electrode (located on the z-axis) and the current is extracted throughout the rest o f the sphere. The thin line represents the current density distribution that is equal to kl=Jo on the su$ace of the electrode and equal to l-cosee k, = J, on the rest of the I + cos ee su$ace. The angle named 0 (as de$ned in figure 2.a) is the angular distance of a measuring point on the surface of the sphere to the injection electrode.
In practice, there are two current injection electrodes at arbitrary positions on the surface of the sphere and two potential measurement electrodes. Because of the linear relation between current and potential (eq. ( 3 ) ) , the overall potential distribution, which results from an injecting and extracting electrode (respectively located on I, and$*), and measured on ? with respect to a reference ) can be defined as: (located on Imr

Y axis

Z-aris

Injected current pattem

In this method the Ip, which consists in the determination of the electrical conductivities given the generated potential distribution originated by a known injected electrical current, was solved through the minimisation of the cost function:

Figure 2. a) Schematic representation of 8 and 9 angle. The angle 8 is the


angular distance of a point P on the surface of a sphere of radius R to the Zaxis, and 9 is the angular distance to the X-axis. b) The angle named 0, is the angular dimension of the electrodes. In this case, there is only one injection 0-7803-6465-1/001$10.0002000 IEEE
Where K k are the measured (or simulated potentials) and Kk((3) the potentials predicted by the model on the i-th potential measurement electrode pair, caused by the k-th current injection electrode pair.
2347

Proceedings of the 22"dAnnual EMBS International Conference, July 23-28, :2000, Chicago IL.

3 .Results
The efficiency of the proposed EIT method was studied in several simulations. The results regarding the robustness of the implemented algorithm to noise proved to be very encouraging, as can be observed in table 1. In fact, even for the lowest signal-to-noise ratio (SNR). conductivity values within 12% of the true values were obtained. Also, it is visible the efficiency of the cost function in improving the estimations with the increase of the SNR values.
01

The ability of the presented EIT to correct for conductivity parameter errors was also studied in a series of simulations in which the EEG inverse problem was solved with systematic conductivity errors, but the correct geometry (figure 4 ) . In all cases the true values for c1 and c3were assumed to be differeat. The EEG inverse problem was solved and a comparison was made between the dipole position and strength errors obtained in two ways:
1. The conductivities used to find the inverse

U2

SNR 10 20

Ed%)

EX%)

30
50

70 100
true values

11.67 1.21 0.22 0.01 0.02 0.09 2.600E-04

7.87 0.02 0.38 0.07 0.07 0.15 5.69E-05

solution were set equal to the true values (ol,02,03), affected by a 50% error; 2. The conductivities, used to find the EEG inverse solution, were given by the EIT method, in the approximation 01=(~3. In the plot of figure 4 it j s represented the error in the dipole position when using: EIT given conductivities in the a.pproximation 0 1 = 0 3 (EIT), (q, 02+50%, cr3) (Case l), (cl, 02-50%. 03)(Case 2), (a1+50%, az-50%, 03+50%) (Case 3) and (01-50%, 02+50%, 0350%)(Case 4).

One of the ideas supporting this method is to use the same head model in both the EIT estimation of the conductivities and in the EEG IP. It was observed that if the head model was affected by some geometrical error then the estimated electrical conductivities would also be affected by an error which would tend to compensate for the geometrical error. For instance, if for the skull thickness a value was used which was twice as large as the true value, its conductivity estimated by EIT would also be twice as large as the true value, and vice versa. Therefore, when using both the head geometry with errors and the EIT estimated conductivies in the solution of the EEG IP, a decrease in the systematic errors could be expected. To test this hypothesis several simulations were performed in which a comparison was made between the case that a dipole was estimated using the wrong skull thickness but the correct skull conductivity and the case where the erroneous skull thickness was compensated using the EIT determined skull conductivity. This comparison was made for various skull thickness errors (expressed in factors of the true thickness) and various ratios of o1and oz.The results are plotted in figure 3.
0-7803-6465-11001$10.00 02000 IEEE

The results are presented in figure 4 .


Effect of geometric errors on diipole position

$
5

0.90
0.80 0.70 0.60

4 0.50

2 c 8

+s

Igralbl@=l

0.40

0.30
0.20

0.10

trlgmlbl@=77

0.00 0.00

0.50

1.00

1.50

2.00

Factor affectlng true "skull thickness"

Figure 3. See text.

2348

Proceedings of the 22 Annual EMBS International Conference, July 23-28,2000, Chicago IL.

II
g
0

.
1.4 I

Effect of conductivity errors on dipole position

References
0 sigma%igmal=1.121 W si~ma3sigmal=2.000 W sigmaWsigmal=3.000

- l2 F
1.0

X 0.6
B
(D

0 . 8

[ll Z.J Koles, Trends in EEG source localization Electroenceph. Clin. Neurophysiol, vol. 106, pp. 127-37, 1998. [21 M. Hihalainen, R. Hari, J. Risto, J.
Knuutila,O.Lounasmaa,Magnetoencephalog
phy - theory, instmmentation, and applications to noninvasive studies of the working human brain, Reviews of Modern Physics, vol. 65, pp. 413-97, 1993. [3] J.C. de Munck, M.J. Peters, A fast method to compute the potential in the multi sphere model IEEE, Trans. Biomed. Eng., vol. BME40(1 l), pp. 1166-1174, 1993. [4] R.M. Leahy, J.C. Mosher, M.E. Spencer, M.X. Huang, J.D. Lewine, A study of dipole localization accuracy for MEG and EEG using a human skull phantom, Electroenceph. Clin. Neurophysiol., vol. 107, pp. 159-73, 1998. [5] J.C. Mosher, M.E. Spencer, R.M. Leahy, P.S. Lewis, Error bounds for EEG and MEG dipole source localization, Electroenceph. Clin. Neurophysiol.,vol. 86, pp. 303-21, 1993. [6] R. Pohlmeier, H. Buchner, A. Knoll, R. Beckmann, J. Pesh, The Influence of Skull Conductivity Misspecification on Inverse Source Localization in Realistically Shaped Finite Element Models Brain Topography, vol. 9 N03, pp. 157-62, 1997. [7] C.J. Stok, The influence of model parameters on EEG/MEG single dipole source estimation, IEEE Trans. Biomed. Eng., vol. 34, pp. 289-96, 1987. [8] T J C Faes, H A van der Meij, J C de Munck and R M Heethaar, The electric resistivity of human tissues (100 Hz-10 MHz): a meta-analysis of review studies Physiological Measurement, vol. 20(4), pp. RI-R10, 1999. [9] L.A. Geddes, L.E. Baker, The specific resistance of biological material - A compendium of data for the biomedical engineer and physiologist, Med. & Biol. Eng., vol. 5, pp. 271-293, 1967. [lo] D.C. Barber and B.H. Brown, Applied potential tomography, J. Phys. E.: Sci Instrum., vol. 37, pp. 723-732, 1984. [ l l ] S . Gonqalves, J.C. de Munck, R.M. Heethaar, F.H. Lopes da Silva, B.W. van Dijk, The Application of Electric Impedance Tomography to reduce systematic errors in the EEG Inverse Problem - A Simulation Study, Physiol. Meas., 1999, submitted.

f 0.4

I !

0 . 2 0 . 0

Er

Case1 Case2 C a s e 3 Case4

slhratlon

Figure 4. See text.

4. Conclusions
This study demonstrates the theoretical feasibility of the proposed BIT method to estimate the equivalent electrical conductivities of brain ( G ~ ) ,skull (q) and On the other hand, it clearly shows scalp (0,). that the presented method has the ability to compute equivalent electrical conductivities which compensate for errors committed on the geometry of the head. The use of the EIT estimated conductivities in the solution of the EEG IP with a wrong head geometry proved to be effective in the decrease of the systematic errors of the dipole position. Also the use of EIT estimated conductivities in the solution of the EEG IP instead of conductivities affected with errors also improves, in general, the dipole position error. It was concluded (data not shown) that the method is not efficient in improving the dipole strength error. It is therefore demonstrated that the combination of EIT and EEG has the potential to reduce systematic errors in estimating the underlying generators of the EEG. More details supporting this conclusion are presented in [ 111. Acknowledgements The work of S . Gonqalves on this project was financially supported by a PhD. Scholarship (Praxis XXI/BD/15502/96) awarded by the Portuguese Foundation for Science and Technology.

0-7803-6465-1/00/$10.00 02000 IEEE

2349

You might also like