Influence of head models on EEG simulations and inverse source localizations
© Ramon et al; licensee BioMed Central Ltd. 2006
Received: 07 October 2005
Accepted: 08 February 2006
Published: 08 February 2006
Skip to main content
© Ramon et al; licensee BioMed Central Ltd. 2006
Received: 07 October 2005
Accepted: 08 February 2006
Published: 08 February 2006
The structure of the anatomical surfaces, e.g., CSF and gray and white matter, could severely influence the flow of volume currents in a head model. This, in turn, will also influence the scalp potentials and the inverse source localizations. This was examined in detail with four different human head models.
Four finite element head models constructed from segmented MR images of an adult male subject were used for this study. These models were: (1) Model 1: full model with eleven tissues that included detailed structure of the scalp, hard and soft skull bone, CSF, gray and white matter and other prominent tissues, (2) the Model 2 was derived from the Model 1 in which the conductivity of gray matter was set equal to the white matter, i.e., a ten tissue-type model, (3) the Model 3 was derived from the Model 1 in which the conductivities of gray matter and CSF were set equal to the white matter, i.e., a nine tissue-type model, (4) the Model 4 consisted of scalp, hard skull bone, CSF, gray and white matter, i.e., a five tissue-type model. How model complexity influences the EEG source localizations was also studied with the above four finite element models of the head. The lead fields and scalp potentials due to dipolar sources in the motor cortex were computed for all four models. The inverse source localizations were performed with an exhaustive search pattern in the motor cortex area. The inverse analysis was performed by adding uncorrelated Gaussian noise to the scalp potentials to achieve a signal to noise ratio (SNR) of -10 to 30 dB. The Model 1 was used as a reference model.
The reference model, as expected, performed the best. The Model 3, which did not have the CSF layer, performed the worst. The mean source localization errors (MLEs) of the Model 3 were larger than the Model 1 or 2. The scalp potentials were also most affected by the lack of CSF geometry in the Model 3. The MLEs for the Model 4 were also larger than the Model 1 and 2. The Model 4 and the Model 3 had similar MLEs in the SNR range of -10 dB to 0 dB. However, in the SNR range of 5 dB to 30 dB, the Model 4 has lower MLEs as compared with the Model 3.
These results indicate that the complexity of head models strongly influences the scalp potentials and the inverse source localizations. A more complex head model performs better in inverse source localizations as compared to a model with lesser tissue surfaces. The CSF layer plays an important role in modifying the scalp potentials and also influences the inverse source localizations. In summary, for best results one needs to have highly heterogeneous models of the head for accurate simulations of scalp potentials and for inverse source localizations.
Highly heterogeneous finite element method (FEM) models of the head have recently become increasingly popular for EEG (electroencephalography) simulations and inverse reconstructions of the electrical sources in the cortex. How does the complexity of these models influence the forward and inverse simulations? We have examined this question with four different FEM models of the head varying in complexities from five to eleven tissue-types. In particular, we examined the effects of CSF, gray and white matter on the forward and inverse simulations for the sources located in the motor cortex area. Our results show that both the scalp potentials and the inverse source reconstruction are significantly influenced by the model complexity.
Previous studies with boundary element method (BEM) models of the head have examined how volume currents affect the forward EEG simulations and also their effects on inverse source localizations [1, 2]. It was found that a 3-compartment BEM model of the head performed better than a 3-shell spherical model of the head, particularly in basal brain areas, including the temporal lobe . Recently, a five tissue-type FEM model of the head has also been used for MEG (magnetoencephalography) simulations and source reconstructions . That study compared the performance of a five tissue-type FEM model with a spherical head model and found that the five tissue-type FEM model performed better in accounting of the volume currents and in inverse source localization. These previous studies show that more complex head models account for volume currents more precisely as compared to simpler, e.g., spherical, head models. Thus, highly heterogeneous finite element models of the head have a potential to further improve the inverse source localizations. In related studies, a five tissue-type FEM model of the head has also been used for efficient computations of the lead fields [4, 5] and also for analyzing the effects of tissue conductivities on MEG forward and inverse simulations .
For simulation studies, four models were used:
Model 1: Full model with eleven tissue-types,
Model 2: Full model with the conductivity of gray matter equal to white matter, i.e., a ten tissue-type model,
Model 3: Full model with the conductivities of gray matter and CSF equal to the white matter, i.e., a nine tissue-type model,
Model 4: Five tissue model consisting of scalp, hard skull, CSF, gray and white matter.
The eleven tissues used in the Model 1 are: scalp, fat, muscle, hard skull bone, soft skull bone, gray matter, white matter, eyes, cerebellum, cerebrospinal fluid (CSF) and soft tissue.
Head tissue resistivity and conductivity values.
Resistivity (Ohm cm)
Brain White Matter
Brain Gray Matter
Spinal Cord and Cerebellum
Cerebrospinal Fluid (CSF)
For the inverse source localizations, first the forward problem was solved. The EEGs were simulated at 145 scalp electrodes using the Model 1. A set of 716 trial inverse runs was made covering the 3 cm cube motor cortex area. For each trial, a dipolar source with random magnitude was placed at a given position in the motor cortex and the scalp potentials were computed. Uncorrelated Gaussian noise was added to achieve the desired signal to noise (SNR) ratio. The SNR was defined as :
where var(V exact ) is the variance of the simulated noisefree observations, and σ2 is the variance of the added noise.
These forward simulated EEGs were then used for inverse source localizations using the lead fields of four different models. The Model 1 was used as a reference model. Inversions were performed with the least-squares technique. An exhaustive search pattern was used, i.e., inversion was performed for each possible source location in the motor cortex and the site producing the smallest residual norm was selected as the best possible source location.
All computations were performed on an Intel 3.2 GHz workstation with 1.2 gigabytes memory. Each run for the lead field computation took between 2–3 seconds. Post-processing and visualizations were done using the Matlab software, version 7.1 (Mathworks, Inc., Natick, MA).
All models have similar MLEs for SNR of -10 to 0 dB. After that differences in MLEs for different models begin to emerge. For all three dipole orientations, Model 1 and Model 2 have the lowest errors in the SNR range of 0 to 30 dB. The Model 3 behaves in a peculiar fashion. MLEs decrease with increasing SNR from -10 to 5 dB. After that for all three dipole orientations the MLEs slightly increase with increasing SNR in the range of 3 to 30 dB. This could be because the Model 3 does not have any morphological distinction between the CSF, gray and white matter and CSF plays an important role in redistributing the volume currents . Errors, i.e., MLEs for the Model 4 are higher as compared to the Model 1 or Model 2. This could be due to the lack of muscle, fat and the skull bone anisotropy in the Model 4 which are included in the Model 1 or Model 2.
Results of the inversions performed with the combined lead fields of x, y and z oriented dipoles are given in Figure 14. This is not an average of the results shown in Figures 11, 12, and 13. These results were obtained by using the combined lead fields of all three, x, y and z orientations of the dipoles. Here the Model 3 has consistently largest error as compared to other models for SNR values in the range of 0 to 30 dB. This is different as compared to the behavior of the Model 3 in Figures 11, 12, 13. The MLEs for the Model 4 are still larger than the Model 1 or the Model 2.
In all of the inversion results given in Figures 11, 12, 13, 14, the STD values are in the range of 6.5 to 5.4 mm at the SNR of -30 dB. The STD values decrease for SNR values from -30 to 0 dB. The behavior of all four models is very different for the SNR values in the range of 0 to 30 dB. For the x oriented dipoles (Figure 11), the STD values for the Model 3 slightly increase after the SNR of 5 dB. The same pattern is also present for the z oriented dipoles (Figure 12). The Model 4 has higher STD values as compared with the Model 1 or Model 2.
These results suggest that head model complexities influence both the forward EEG simulations and the inverse source localizations. The above results also suggest that the Model 3 has larger source localization errors as compared to the full model, i.e., Model 1. In Model 3, the difference between the CSF and brain matter was eliminated. The resistivity of the CSF is less than the gray and white matter. So in the full model, i.e., Model 1, the currents will follow the structural paths of CSF channels in the brain. In Model 3 this distinction does not exist and the spread of the currents will be more uniform as compared to the Model 1 and 2. This will change the scalp potentials over a large portion of the scalp surface. It has been shown earlier that CSF plays an important role in distributing the currents in a head model . This could also be the reason that Model 3 performs much worse in inverse source localizations as compared to Model 1 or Model 2. The electrical conductivity of the human CSF is well documented in the literature  and can be incorporated in head models.
These model dependant results should also be compared with the tissue conductivity related results where one changes the tissue conductivity in steps and examines the changes in the scalp potentials [19–22]. Previous studies have not eliminated tissue boundaries, but they have used incremental changes in the tissue conductivities or have used upper and lower bounds of the tissue conductivity values [6, 21]. Also, detailed scalp potential maps are not available in previous studies to perform a comparative analysis. In general, previous studies have found that both the forward and inverse results are severely influenced by changes in the conductivity of skull bone, CSF, gray and white matter. In particular, conductivity of skull bone [6, 21–23] and the skull anisotropy  severely influences the EEG and MEG simulations and inverse source reconstructions. Conductivity related inverse localization errors could be of the order of 2.35 mm to 9.59 mm . Our results also show that localization errors increase as the complexity of the model decreases. The fat, muscle and soft bone structures are not included in the Model 4 and this model has larger source localization errors as compared to the Model 1 or Model 2. This suggests that highly heterogeneous finite element models of the head are needed to reduce the source localization errors. Our work here was limited to dipoles in the motor cortex area. However, one could expect similar results for dipoles located in other parts of the cortex.
As our results show that CSF layer plays an important role. It influences simulations of scalp potentials and also influences the inverse source localizations. The CSF layer is difficult to segment in T1 or T2 weighted MR images. One needs to be aware that miss-segmentation of the CSF layer could become a source of error in EEG computations. There is also a related issue regarding the position of the brain and the CSF layer. The MRI data is collected while the subject is in a supine position and the EEG data is collected while the subject is, generally, in a sitting position. There could be a slight difference in the location of the brain within the skull when a subject is in supine position as compared to when the subject is sitting. The CSF layer could also be slightly shifted due to a shift in the brain position. This shift in the brain position between the supine and sitting position is an additional source of error which is difficult to account for while building the computer models of the head.
Here we have used an exhaustive search pattern to localize the sources. This means that all the possible nodes were searched in the 3 cm cubic volume of the motor cortex. The node producing the least error was selected as the possible source location. This provides the best behavior of a given model in inverse source localizations. The reported mean localization errors are the best results one could expect from a given model.
This work was supported in part by the National Science Foundation under Grant No. 0112742
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.