Vous êtes sur la page 1sur 3

Proc.14th Ann.Int.Conf.IEEE Eng.Med.Biol.Soc.,Satellite Symp. Lyon, 1992, pp.

3436

Magnetic Localization of EEG Electrodes for Simultaneous EEG and MEG Measurements

Wolfgang Becker, Volker Diekmann and Reinhart Jrgens


Sektion Neurophysiologie, Abteilung Neurologie Universitt 7900 ULM GERMANY

Abstract We describe a type of electrode with an embedded magnetic coil designed for concurrent EEG recording during MEG measurements. By means of the embedded coils and a least squares fitting algorithm the location of these electrodes can be specified in both head and cryostat based coordinate systems. The electrodes have a height of only 4 mm and therefore do not hinder cryostat positioning close to the head.

Rsum

I. INTRODUCTION

wires that run to a threecontact connector which also receives the lead from the pellet. Computer simulations show that the

An ultimate goal in magnetoencephalography is source localization on the basis of simultaneous magnetic (MEG) and electric (EEG) measurements. This requires that the positions of both the MEG sensors and EEG electrodes be accurately known in relation to a head based reference system. Systems that measure the position of triplets of orthogonal coils attached to landmarks on the head with respect to an array of sensors have been described recently, e.g. [1], [2]. In principle, if such triplets were fixed on top of EEG electrodes, the position of these electrodes could determined by the same method. However, from a practical point of view this is an unfeasible approach since the triplets, which typically are supported by plastic cubes of at least 1 cm3, would increase the spacing between head and cryostat and increase the danger of accidentally pulling off electrodes, in particular in patients. We therefore have developed extra flat EEG electrodes with an embedded magnetic coil as well as computer programs that localize the position of these coils and determine the orientation of their axis. With the electrodes sitting tangentially to the head surface, and hence the coil axes oriented perpendicularly, also information about head shape can be derived. In the following we shall first describe the construction of the electrodes as well as the principles of data acquisition and analysis underlying localization, and then report first experiences with an implementation of these procedures in a 28channel magnetometer system.
II. ELECTRODES

4mm 10mm 16mm

The electrodes consist of custommade AgAgCl sinter pellets (disks of 12 mm diameter, 1 mm height) with a braided silver wire entering across the rim for contact. To one face of the electrode is glued a spiral coil (Fig.1B) etched on a thin (0.25 mm) substrate (Duroid; Rogers, Chandler, AZ) and electrically insulated by a varnish. The spiral has 10 turns with a radius increasing by 0.03 mm per turn, starting from an initial value of 2.7 mm. Its start and end tabs are connected to a pair of closely twisted

5.7mm 11.2mm

Fig. 1. EEG electrode with embedded coil. Cross section of electrode (A) and layout of coil (B).

Proc.14th Ann.Int.Conf.IEEE Eng.Med.Biol.Soc.,Satellite Symp. Lyon, 1992, pp. 3436


field of the spiral closely matches that of a set of concentric circular coils; deviations from circular symmetry occur mainly in the vicinity of the lead insertions and become negligible at distances of more than 1 mm. The pelletcoil assembly is protected and mechanically secured by an encasement, molded from polyurethane (Dekorit F; Raschig GmbH, Ludwigshafen), with a 10 mm diameter opening through which the electrodeskin contact is established (Fig 1A). The readytouse electrodes have a diameter of 16 mm and total height of only 4 mm. The electrodes are fixed to the head by a gluing electrode cream (EC2; Grass, Quincy, MA), but collodion may be used as well since the encasement is solvent resistant.
III.

DATA ACQUISITION

The data required to localize a coil are obtained by energizing the coil, under computer control, by a sinusoidal current (8 Hz) with an initial magnitude of 1 mA. The field created by this current is sensed by an array of magnetometers the output of which is digitized (sampling rate 500 Hz), stored in memory, and visualized on a monitor screen. The field magnitude induced in each channel of the array then is calculated by extracting the fundamental component contained in the corresponding signal (Fourier analysis across 7 cycles of the signal). If in any of the arrays channels the field exceeds 10 nT, the measurement will be repeated with a lower current to prevent an overload of the channels. If, on the other hand, none of the channels sees a field of at least 2.5 nT, the measurement is repeated with a correspondingly increased current in order to optimize the signaltonoise ratio. Once a satisfactory current setting has been found, the distortion factors (obtained from the Fourier analysis) of all channels with sizeable signal amplitude (> 5 pT per mA of coil current) are examined. Large distortion factors (> 5 %) are taken as warnings of possible flaws in the measurement such as may be caused by unusually large surges of the extraneous field or by movements of the head (and hence of the coil) with respect to the sensor array. In such cases the program will prompt the intervention of the operator who may decide to repeat the measurement or to reject the results of particular channels. Finally, the set of accepted results, [aij] (amplitude induced in sensor i by coil in electrode j), is stored for later processing , and the program proceeds to apply the same procedure to the coil of the next electrode.
IV. LOCALIZATION

tions show that for all realistic coiltosensor distances (>15 mm or so, because of the intervening cryostat wall), the gain in accuracy from an exact integration of BiotSavarts law along the spiral would be less than 0.2 mm, at the cost of an incommensurate increase in processing time. Knowing electrode position in relation to the sensor array (cryostat based coordinate system) is not enough; what one is interested in is electrode position in relation to a head based coordinate system, e.g., the one defined by the nasion and the two preauricular points [1]. We therefore determine also the positions of these landmarks with respect to the sensor array, using triplets of orthogonal coils which, at these locations, do not interfere with the cryostats; triplet localization follows the ideas outlined in [1] and [2], and uses a computer controlled optimization of data acquisition similar to that described above for electrode coils. From the positions of the three landmarks the transformation is derived by which electrode positions can be related to the head coordinate system. Since head position may change with respect to the sensor array in the course of an investigation, the landmarks must be surveyed repeatedly in order to keep track of the the head vs. array position which is important for the interpretation of the magnetic signals. In contrast, electrodes need be surveyed only once since, unless pulled off, they do not change their position on the head.
V. RESULTS OBTAINED WITH IMPLEMENTATION OF ELECTRODE LOCALIZATION INTO A DORNIER MULTICHANNEL MEG SYSTEM

PROCEDURE

Given the set [aij], the position and orientation of electrode j in relation to the sensing array is determined by fitting a model coil such that

(a
i

ij

* A i) 2 + min

(Ai, field of model coil at position of sensor i); the fitting procedure is based on a Marquardt algorithm [3] (and draws on an earlier version developed by H. Effinger of Dornier GmbH, Friedrichshafen). As a model coil we use an approximation of the spiral current by a set of concentric circular currents; simula-

Our system of electrode localization has been incorporated into a 28channel magnetometer system (Dornier GmbH, Friedrichshafen) comprising two cryostats with 14 channels each operating in a shielded room (Vacuumschmelze GmbH, Hanau); it is now being used routinely with epileptic patients. Although the two cryostats of this system can be positioned independently of each other, for any given configuration of cryostat positions, mutual magnetic localization of one cryostat by its companion one allows the construction of a common reference system (method provided by Dornier). Another feature of the instrument, the use of a software gradiometer to suppress extraneous noise, is needed only when recording biological data; when localizing sources of high signal to noise ratio such as our electrode coils, the original magnetometer signals can be used. Typically, patients are equipped with 19 coil electrodes distributed over the cranium according to the tentwentysystem [4] plus reference electrodes without coils. After a patient has been seated under the cryostats and his head has been stabilized, the program first switches in the triplets to record the data required for establishing the relation between the head and cryostat coordinate systems. Thereafter the nineteen electrode coils are consecutively energized and the set of coefficients [aij] is calculated. The two procedures together take less than 2 min. Subsequently both the MEG and EEG are recorded; so far, no MEG artifacts have been observed that could be attributed to the electrodes attached to the patients heads. Whenever head movements are suspected, the head vs. cryostat calibration is rerun

Proc.14th Ann.Int.Conf.IEEE Eng.Med.Biol.Soc.,Satellite Symp. Lyon, 1992, pp. 3436


which takes less than 30 sec. The least squares fit algorithm used to extract the various position informations is run offline at the conclusion of the patients examination. To increase the chance for the the algorithm to converge and to do so within reasonable time, it is important to provide appropriate start values. Using a set of initial values reflecting the rules of the tentwenty system, we have obtained convergence in all of the 110 electrode positions from 6 patients investigated so far. To obtain an indication of the methods accuracy under realistic conditions we have measured the actual electrode spacings with a caliper in these patients. Fig. 2 gives a frequency histogram of the differences between caliper and magnetic measurements; it will be appreciated that the majority of these differences were smaller than 4 mm and that 50% were even less than 2 mm. Similar results were obtained when only a single cryostat (14 channels) was operated, although the tolerance toward occasional malfunctions of single channels and towards small calibration errors of the system decreased. Not unexpectedly, a closer analysis shows that the largest errors occur for electrodes which are outside the area covered by the sensor array (which sees only about 40 % of the surface over which electrodes are distributed). This was particularly true for electrodes in the occipital area which, besides being outside the arrays field of view, were in most patients also not normally oriented with respect to the pickup coils of the array. Obviously, however, these errors do not represent intrinsic flaws of the above method; at present they can be overcome by repositioning the cryostats, and they will disappear with the introduction of helmet type MEG systems in a near future. REFERENCES
[1] S. Ahlfors and R.J. Ilmoniemi, A magnetometer position indicator for multichannel MEG measurements, in Advances in Biomagnetism, S.J. Williamson, M. Hoke, G. Stroink and M.Kotani, Eds. New York: Plenum, 1989, pp. 693696. [2] D.B. van Hulsteyn, L. McGavran, L. Heller, J. George, P. Medvick and E. Flynn, in Advances in Biomagnetism, S.J. Williamson, M. Hoke, G. Stroink and M.Kotani, Eds. New York: Plenum, 1989, pp. 689692. 1 3 5 7 9 11 13 15 17 19 21 23 25 [3] D.M. Marquardt, An algorithm for least squaresestimation of nonlinear parameters, J. Soc. Indust. Appl. Math., vol.11, pp. 431441, 1963. [4] H.H. Jasper, The tentwenty electrode system of the International Federation, Electroencephalogr. Clin. Neurophysiol., vol. 10, pp. 371375, 1958.

50 Frequency of occurrence (%) 40 30 20 10 0 Midpoint of error class (mm)

Fig. 2. Frequency histogram of errors between magnetically determined electrode distances and those obtained with caliper.

Vous aimerez peut-être aussi