Electroencephalography (EEG) is an indispensable neurological diagnostic tool in terms of the fast time scale, porta-bility and cost efficiency. Improved spatial res-olution of EEG measures would greatly bene-fit multiple clinical and research applications, including stroke, epilepsy and cognitive stud-ies. The recent advances in dense arrays elec-trode application have made feasible EEG brain imaging for both rapid application and long-term monitoring.1 It has been shown that reliable inverse solutions can be obtained and dense array sampling (128, 256 and 512 chan-nels) on the scalp can be projected back to the cortex providing a unique opportunity for monitoring brain activity both in space and time (Figure 1). However, the spatial accuracy of EEG will remain limited because i) mostly simplistic models of the human head (like multi-shell spheres) are commonly used in the inverse procedure of back-to-cortex projec-tion, and ii) the regional conductivities of the human head tissues are largely unknown. This is true not only in each measurement case but in general. Several imaging modalities have been proposed so far to quantitatively measure the electrical conductivity of tissue noninva-sively, but none of them is free from some lim-itations and shortcomings. Magnetoacoustic Hall effect imaging2 relies on propagation of ultrasound into the tissue, and is not quantita-tive.
Magnetic resonance current density imaging3 requires applying rather high level of external currents to make produced magnetic field contrast visible by MRI. The electrical conductivity tensor of tissue can be quantita-tively inferred from the water self-diffusion tensor as measured by diffusion tensor mag-netic resonance imaging (DTI).4 It can be suc-cessful in extracting anisotropic conductivities of the brain tissue, but more problematic regards bone (skull) tissues where the water content is much smaller. The lack of accurate skull conductivity (most resistive tissue) is particularly problem-atic given the developmental variations in the human skull from infancy through adoles-cence. Without an accurate forward model of the skull (specifying the volume conduction from cortex to scalp) even advanced inverse efforts cannot achieve precision with EEG data as the error of source localisation due to conductivity uncertainty may reach a few cen-timetres.5 Several authors addressed this prob-lem by using the noninvasive electrical impedance tomography (EIT) approach. A similar approach was used by Hoekema et al6 in the semi in-vivo conductivity measure-ments of the skull parts temporarily removed during epilepsy surgery, - fitting for only one unknown parameter was performed, though. Goncalves et al7,8 applied spherical and a three-layer boundary element (BEM) models to fit their EIT measurements for six subjects.9 However, since in such models skull thickness and conduc-tivity are interchangeable to some extent, more accurate geometry representation is needed. Recently we have shown in our group10,11 that using the parameterised EIT measurements procedure and realistically shaped high-resolution finite dif-ference models (FDM) of the human head based on the subject specific co-registered CT and MRI scans, as is shown in Figure 2, it was possible to extract three and four tissues conductivities (Table 1) both in simulations with synthetic and real experimental data with good accuracy using the multi-start downhill simplex algorithm.
Our current focus is to further improve con-ductivity information for the benefit of high-res-olution EEG source localisation. We are conduct-ing measurements repeatedly on individual sub-jects to prove the method’s robustness and show individual variability of head tissues conductivity across individuals. Electrical Geodesics, Inc. (EGI) has developed a data acquisition system that pro-vides current injection between selected electrode pairs (at very safe current levels) and simultane-ous acquisition of return potentials from the dense sensor array of the Geodesic Sensor Net. Our current work will further refine a methodol-ogy for constructing accurate forward models of electrical conductance for the human head through incorporating the high-resolution struc-tural details of the human head from MRI/CT scans and providing the non-invasive procedure for estimation of the major tissues conductivity parameters. The latter is based on parameterised EIT inverse solution for the data collected at EGI. In the next stage of the project, a refined forward solver will incorporate anisotropies of the head tissues, in particular skull and white matter.
The advanced simulation annealing algorithm has been proved to show better performance in the inverse procedure in terms of finding the global minima of the cost function with larger number of unknowns. This will allow us to extend the pro-cedure to a parcellated skull (10-12 anatomically relevant bone plates) and include the skull con-ductivity inhomogeneities information into the forward solver for the EEG inverse problem. These tasks require extensive computational resources. At Neuroinformatics Center, University of Oregon we have access to a multi-cluster (SGI, Dell, IBM p650; IBM p690; IBM BladeServer) high performance parallel computing system dedicated to analysis of human EEG and MEG data. The first and primary application of these computing resources is to solve the conductivity problem of the human head.