A study of dipole localization accuracy for MEG and EEG using a human skull phantom
Introduction
Electroencephalography (EEG) and magnetoencephalography (MEG) can be used, respectively, to measure scalp surface potentials and external magnetic fields produced by the neural current sources associated with sensory, motor and cognitive activity. Inverse procedures in EEG and MEG are used to estimate the spatial distribution of the underlying, possibly focal, neural sources. The equivalent current dipole, and clusters of such dipoles, are a widely used source model for representing focal neural activity. For this model the inverse procedure must estimate the locations and amplitudes of the equivalent dipoles.
An important step in assessing the accuracy with which these sources can be estimated is to perform experimental studies in which the true location and temporal activity of the dipoles are known. In this way we can study the effect on accuracy of errors in the head and sensor models and of noise in the data. Studies of this type can be performed using computer simulation; however, the majority of published results that use computer simulations assume simplified models for the head, instrumentation and noise. Typical simulations use a spherical head with point measurements of the scalp potential or magnetic field and additive white Gaussian noise (cf. Mosher et al., 1993). To establish practical limits on the accuracy with which dipolar sources can be estimated, the model should take into account the non-ideal nature of the sensors, realistic head geometries and correlations in the noise. Furthermore, we must also consider the effects of inaccuracies in the forward model associated with uncertainties in the estimated conductivities in the head, and the effect of simplifications and numerical errors associated with either spherical head approximations or boundary element methods based on more realistic head geometries.
While more elaborate simulations could be developed to include these factors, an evaluation using data collected directly from a physical system has the advantage that the results can more closely reflect in vivo performance since they include factors that cannot readily be included in simulations such as environmental noise and deviations of the physical system from our model. Such studies have been performed using dipolar sources implanted in epilepsy patients (Cohen et al., 1990); however, the procedures required to implant these sources, including making holes in the skull, may result in severe distortion of volume currents. While such studies are important, they are not ideal for evaluation of general forward and inverse methods. The other functional modalities (fMRI and PET) offer the potential for providing ground truth for clinical and volunteer studies; however, the relationship between hemodynamic and electrophysiological processes are currently not sufficiently well understood to provide reliable cross-validation.
A multiple dipole phantom was used by Phillips et al. (1997)for evaluation of several MEG imaging methods. This `dry' phantom is based on the theoretical description of Ilmoniemi et al. (1985)in which the resulting fields are shown to be identical to those produced by a current dipole in a uniformly conducting medium. The major limitations of this phantom are its inability to generate the volume currents associated with realistic head geometries and its unsuitability for EEG. Interesting studies have been performed with dipoles implanted in a cadaver head (Barth et al., 1986) and gelatin filled skulls by Greenblatt and Robinson (1994)and Lewine et al. (1995), and more recently by Baillet et al. (1998). Here we build on these studies using a large number of dipoles implanted in a human skull phantom.
Motivated by the desire to produce realistic data corresponding to complex spatio-temporal current sources and to include the effects of realistic head geometries, we designed and fabricated a multiple dipole phantom consisting of 32 independently programmable and isolated dipoles which can be inserted in a skull mount and used to collect both EEG and MEG data. The design of the phantom was first described by Spencer et al. (in press). Here we report on the design of this phantom and include the results of a localization study using EEG and MEG. The design of the phantom and data collection procedures are described in Section 2. In Section 3we describe our data processing methods that include registration of the MEG and phantom-based coordinate systems, forward head modeling for EEG and MEG, and current dipole fitting. Experimental results are reported in Section 4. These include tabulations of the localization errors for each of the dipoles using EEG and MEG data with spherical and realistic head models. In Section 5we include the results of several simulations that are compared with the experimental data in order to assess the impact of different error sources on the total localization error. Final conclusions are drawn in Section 6.
Section snippets
Phantom design
The phantom design consists of 3 components: (i) a 32-element current dipole array; (ii) a personal computer (PC) controlled dipole driver with 32 isolated channels allowing independent control of each dipole; and (iii) a human-skull mount in which the dipole array is placed. We describe each of these 3 components below.
Data registration
Ordinarily, only a few fiducial markers are available between different modalities to provide data alignment, but the EEG sensors affixed to the phantom scalp provided a rich set of markers which were used for registration. The MEG and EEG sensor locations were found relative to the PCS using the HPI as described above. The EEG sensor locations were also manually identified and extracted from the CT images. The EEG sensor locations in the PCS, measured using the HPI and the CT identified
The data
The MEG and EEG data corresponding to each of the 32 dipoles were used to compute estimates of the dipole locations, orientations and time series. Before presenting the results of the localization study, we first investigate the signal to noise ratios (SNRs) of the two data sets. In Fig. 6 we have plotted the SNR for each MEG and EEG data set. Here, we define SNR as the root mean square (RMS) value of the measured signal across all the measurement channels and time slices divided by the RMS
Discussion
The results presented above show that MEG dipole localization errors are as small as could be expected (~3 mm) given the various sources of registration error between the true dipole locations, the CT extracted locations and the HPI-based localization of the MEG sensors. In contrast the errors for EEG localization are significantly larger (~7–8 mm), although still well within bounds that could be considered useful in clinical and research localization studies. Here we discuss the factors that
Conclusions
We have presented a study of MEG and EEG dipole localization accuracy using a human skull phantom containing 32 independently programmable and isolated dipolar sources. This phantom provides insights not readily obtainable from either simulation or experimental human data: (i) the skull phantom provides realistic bone structure and morphology; (ii) the EEG and MEG data can be collected on commercial systems, adding the uncertainties of instrumental and environmental noise; (iii) the sources are
Acknowledgements
We would like to thank Bijan Timsari and Tong Zhang of the Signal and Image Processing Institute at the University of Southern California for their assistance in the X-ray CT surface extraction and data registration. Also, we thank Charles C. Wood of the Biophysics Group at Los Alamos National Laboratory for reviewing a draft of this manuscript and providing helpful suggestions. This work is supported by the National Institute of Mental Health Grant RO1-MH53213, and by Los Alamos National
References (22)
- et al.
Magnetic localisation of a dipolar current source implanted in a sphere and a human cranium
Electroenceph. Clin. Neurophysiol.
(1986) - Baillet, S., Marin, G., le Rudulier, F. and Garnero, L. Evoked potentials from a real skull phantom-head: an...
- Brody, D.A., Terry, F.H. and Ideker, R.E. Eccentric dipole in a spherical medium: generalized expression for surface...
- et al.
MEG versus EEG localisation test using implanted sources in the human brain
Ann. Neurol.
(1990) - et al.
A complete linear discretization for calculating the magnetic field using the boundary element method
IEEE Trans. Biomed. Eng.
(1994) - Golub, G.H. and Van Loan, C.F. Matrix Computations. Johns Hopkins University Press, Baltimore, MD,...
- et al.
A simple head shape approximation for the 3 shell model
Brain Topogr.
(1994) - et al.
Realistic conductor geometry model of the human head for interpretation of neuromagnetic data
IEEE Trans. Biomed. Eng.
(1989) - Ilmoniemi, R.J., Hamalainen, M.S. and Knuntila, J. The forward and inverse problems in the spherical model. In: H....
- Lewine, J.D., Edgar, J.C., Repa, K., Paulson, K., Asture, R.S. and Orrison, W.W., Jr. A physical phantom for simulating...
Multiple dipole modeling and localisation from spatio temporal MEG data
IEEE Trans. Biomed. Eng.
Cited by (314)
Realistic three-layer head phantom for optically pumped magnetometer-based magnetoencephalography
2023, Computers in Biology and MedicineElectromagnetic source imaging predicts surgical outcome in children with focal cortical dysplasia
2023, Clinical NeurophysiologyAn in–vivo validation of ESI methods with focal sources
2023, NeuroImageA hybrid steady-state visual evoked response-based brain-computer interface with MEG and EEG
2023, Expert Systems with Applications