EEG forward and inverse modeling in neonates
The ultimate goal of functional brain imaging using EEG is to localize cerebral sources using volume conductor models. The accuracy of EEG source localization largely depends on that of EEG forward modeling, which requires accurate realistic head models incorporating the exact anatomical geometry of head tissues including scalp, skull, cerebrospinal fluid (CSF) and brain. A realistic head model is usually reconstructed through segmentation of individual magnetic resonance images (MRIs). In adults, the main sources of inaccuracies in EEG forward modeling are CSF, inhomogeneities in skull and brain tissues including gray and white matter, and uncertainty in conductivity of different head compartments. In newborn infants, little effort has been made to investigate the effect of uncertainty in head tissue conductivities and geometrical complexities on the accuracy of EEG forward and inverse solutions.
Forward simulations
In a study published in 2016, we conducted a series of simulations by using the finite element method (FEM) to investigate the effect of different neonatal head model deficiencies and inaccuracy in the head tissues' conductivity on EEG forward and inverse modeling. For this purpose, we first segmented coregistered head MR and CT head images from one neonate. We then created a realistic neonatal head model including scalp, skull, fontanels, CSF, gray matter (GM) and white matter (WM) as shown in Fig. 1.
Figure 1. Coronal, sagittal and axial views of the segmented head including GM, WM, CSF, cranial bones, fontanels and scalp, and their corresponding 3D reconstruction.
To investigate the effect of head model inaccuracies and conductivity mismatch on EEG forward modeling in neonates, two head model sets were constructed. The first set was created to investigate the effect of CSF and fontanel exclusion and discrimination between gray and white matters (MoldeT1-3). In the second set (ModelT4 – ModelT11), the effect of variations in conductivity of various compartments was investigated with respect to the reference values derived from the literature as summarized in Table I.
Table I. Summary of the test/reference models and the compartment conductivities (in S/m). WM: white matter, GM: gray matter, CSF: cerebrospinal fluid.
Twelve FEM volume conductor head models were created with tetrahedral meshes containing 633,728 nodes and 3,601,917 elements labeled as scalp, skull, fontanel, CSF, GM, or WM based on the head tissue segmentation. To create a source space, the cortical surface mesh was extracted from the cortical tetrahedral mesh. EEG forward simulations were then performed by placing dipole sources on the vertices of the cortical surface mesh with normal orientation (Fig. 2).
Figure 2. Distribution of dipole sources with normal direction.
To assess differences in the topography and magnitude of scalp potentials calculated for the FEM volume conductor head models, we used the relative difference measure (RDM) and the magnitude difference measure (lnMAG) defined as:
Where N is the number of EEG electrodes and VM1 and VM2 denote the computed scalp potentials with Head Model1 and Head Model2, respectively.
Our results showed that the exclusion of CSF from the head model had a stronge widespread effect on EEG forward solutions. The discrimination between gray matter and white matter also showed a strong effect but less intense than that of the CSF exclusion. The exclusion of the fontanels from the neonatal head model only affected areas beneath the fontanels, but this effect was much less pronounced than that of the CSF exclusion and GM/WM discrimination. Changes in the GM/WM conductivities by 25% with respect to the reference values had considerable effects on EEG forward solutions, with a stronger effect observed for the GM conductivity. Similarly, changes in the skull conductivity affected EEG forward solutions in areas covered by the cranial bones. The least effect on EEG was produced by changes in the conductivity of the fontanels (see following figures). Our findings demonstrated the impact of inaccuracies in conductivity of head tissue compartments on EEG forward modeling in neonates (Azizollahi et al., 2016).
Effect of CSF exclusion
Effect of skull conductivity
Effect of discrimination between gray and white matter
Effect of gray matter conductivity
Effect of exclusion of fontanels
Effect of white matter conductivity
Inverse simulations
In the second study, we conducted a series of numerical simulations to investigate the effect of head modelling errors on EEG source localization in term neonates. We designed three simulation setups to quantify the effect of CSF and fontanel exclusion as well as GM/WM distinction by introducing structural deficiencies into a realistic neonatal volume conductor head model including scalp, cranial bones, and fontanels, CSF, GM and WM.
In the first simulation setup, we investigated the effect of CSF exclusion on EEG source localization. To perform forward and inverse simulations, a reference model (ModelR1) including scalp, skull, CSF, and brain was constructed. In the brain layer, no distinction was made between GM and WM. A test model (ModelT1) similar to the reference model was then created with the exception of the CSF layer, which was excluded from the model by setting its conductivity value to that of the brain layer.
The second simulation setup was designed to demonstrate the effect of GM/WM distinction on EEG inverse solutions. The reference model (ModelR2) used in this simulation was similar to ModelR1 (including scalp, skull, CSF, and brain) except for the GM and WM compartments, which were assigned different conductivity values. The reference model (ModelR1) used in the first simulation setup was used as the test model (ModelT2) in the second simulation setup.
In the third simulation setup, we investigated the effect of the exclusion of fontanels on EEG source analysis in neonates. The reference model (ModelR3) for this simulation was similar to ModelR2 including scalp, cranial bones, CSF, GM and WM except for the fontanels, which were assigned a different conductivity. In the test model (ModelT3) , the skull layer was created by combining the cranial bones and fontanels by attributing the same conductivity value to them. Table II lists the characteristics of the reference and test models for each simulation setup.
Table II. Summary of the test and reference models included in each simulation setup.
We quantified source localization errors in dipole position, orientation and magnitude, caused by differences between the reference and test models in each simulation setup. Our results showed that the exclusion of CSF from the head model could cause significant localization errors mostly for sources closer to the inner surface of the skull. With a less pronounced effect compared to the CSF exclusion, the discrimination between GM and WM also widely affected all sources, especially those located in deeper structures. The exclusion of the fontanels from the head model led to source localization errors for sources located in areas beneath the fontanels. Our results clearly show that the CSF inclusion and GM/WM distinction in EEG inverse modeling can substantially reduce EEG source localization errors. Moreover, fontanels should be included in neonatal head models, particularly in source localization applications, in which sources of interest are located beneath or in the vicinity of fontanels (see following figures). Our finding have practical implication on how head modeling errors can impact the results of EEG source localization in neonates (Azizollahi et al., 2020).
Effect of CSF exclusion
Effect of discrimination between gray and white matter
Effect of exclusion of fontanels