Forward model of rat electroencephalogram: influence of inhomogeneous and anisotropic skull Authors: David Kuratko, Vlastimil Koudelka, Cestmir Vejmola, Filip Dusek, Frantisek Kucera, Daniel K Wójcik, Tomas Palenicek, Jaroslav Lacik Published: 2026 ABSTRACT AbstractThe forward modeling of electroencephalogram (EEG) plays a crucial role in solving source localization and, consequently, in the detection of EEG connectivity networks in the rat brain. We investigated the impact of skull tissue properties—inhomogeneities, anisotropy, and thickness—on measured skull potentials. Five cuboid head phantoms with identical geometry, differing in skull material parameters, and a realistic head phantom with varying skull anisotropy ratios were tested. Simulations were verified by four electromagnetic solvers and an EEG measurement on a physical phantom. The cuboid phantom with sutures yields similar electric potentials as the isotropic one, but noticeable differences appear at electrodes near sutures, particularly in real brain sensing of somatosensory, motor, and visual cortices, as well as the retrosplenial dysgranular cortex, for a dipole located close to the surface. Skull thickness reduces potentials at the top electrodes but increases them at side electrodes, in real brain sensing somatosensory, auditory, association, and entorhinal cortices. Skull anisotropy alters electric potentials with differences up to 60% for dipoles near electrodes at the limiting anisotropy ratio of 10:1, and its effect decreases with increasing distance. Varying anisotropy ratios (10:1 to 1.3:1) show that nearby dipoles are the most affected, while the impact is reduced for lower, more realistic ratios (1.3:1 to 1.8:1) and for more distant sources. This trend was consistently observed in realistic head phantoms as well. Tangential dipoles are most strongly affected in the vicinity of the electrodes and gradually become radial at more distant positions near the inner skull surface.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-58885-1.Subject terms: Engineering, Neuroscience, Physics Introduction Identifying the location of neural activity is essential for the study of brain functions1,2. The task can be achieved through electroencephalogram (EEG) source localization which involves solving both the forward and inverse problems1,3. The forward one evaluates electric potentials on a head surface generated by current sources located inside a volume conductor model of a head1,4. The current dipoles are usually exploited for approximation of those sources since the dominating contribution to the EEG comes from synaptic activation of cortical pyramidal cells generating synchronized currents along their parallel dendrites leading to a dipolar pattern1,5. The volume conductor model of a head contains information about the geometry and electric conductivity of head tissues1,2. In case of a human volume conductor model, the geometry corresponds well to a single or multilayer sphere. However, the accuracy of geometry description has a significant influence on the forward modeling accuracy6–9. Therefore, magnetic resonance imaging (MRI) and computed tomography (CT) are used to describe realistic geometry8–10. The volume conductor model is further sensitive to the correct definition of electric conductivity of individual head compartments7,11 consisting of the brain, the skull, the cerebrospinal fluid (CSF), and the scalp1,4. Accurate EEG source localization in rats is crucial for translational neuroscience research, particularly in understanding brain network dysfunction in neuropsychiatric disorders. Rat subjects enable invasive electrophysiological recordings that complement human non-invasive EEG, bridging the gap between molecular mechanisms and clinical biomarkers12,13. Of particular interest are large-scale brain networks such as the Default Mode Network (DMN), which in rats comprises anterior hubs (prelimbic cortex, cingulate cortex) and posterior hubs (retrosplenial cortex, posterior parietal cortex) organized into functionally distinct subsystems14,15. The rat DMN shows remarkable homology to the human DMN and exhibits altered functional connectivity in animal paradigms of depression, anxiety, and following pharmacological interventions16,17. Additionally, sensorimotor networks, including the primary somatosensory and motor cortices, generate characteristic 8–12 Hz “mu”/alpha-band rhythms during locomotion and encode movement kinematics18,19. Understanding the spatial distribution and orientation of neural sources from these functionally distinct networks is crucial for developing standardized, translational EEG methods comparable to those used in human clinical research. Such methods would enable anatomically precise mapping of electrophysiological biomarkers to specific brain structures in preclinical setups, supporting drug development and mechanistic studies of neuropsychiatric disorders. The forward model exploits a quasi-static approach of Maxwell equations due to low frequencies of brain waves1,2 where the capacitive component of tissue impedance, the inductive effect, and the propagation constant can be neglected1,10. Thus, measured electric potential on the volume conductor surface produced by current sources can be found by solving the Poisson’s Eq. 4, which can be solved for a simple volume conductor such as a sphere and a multi layered sphere semi-analytically22,23. Since the geometric accuracy is crucial, the numerical methods are exploited for more realistic volume conductor models6–9. A variety of numerical methods are utilized for solving forward models, such as the boundary element method (BEM)24,25, the finite difference method (FDM)26,27, and the finite element method (FEM)11,28. The FEM provides the most flexibility in defining geometry and electric conductivity of volume conductor model, however, this is achieved at the cost of the higher computational demands10. On the other hand, BEM and FDM are more limited in terms of geometry and accuracy, but are computationally less demanding1,4. Recently, hybrid techniques such as the boundary element fast multipole method (BEM-FMM)29,30 have been proposed allowing efficient simulations of detailed head geometries with multiple tissues. BEM-FMM enhances the standard BEM formulation, enabling the utilization of significantly finer meshes and more complex geometries while maintaining computational efficiency and providing accuracy comparable to FEM. The skull is composed of three layers, comprising both compact and spongiform bone. These layers differ in electric conductivity resulting in anisotropic behavior31,32. Variation in reported electric conductivity partly arises from different measurement methodologies. The study31 reported lower values obtained by electric impedance tomography (EIT) compared to direct electrical conductivity measurement (DAC) and electro/magnetoencephalography (EMEG). Moreover, factors such as age, pathology, and individual anatomical differences contribute to variability in conductivity measurement31. Variability in electric conductivity with temperature and the frequency of applied electric field has been reported31,33. Moreover, variability in skull conductivity was observed in different pathological conditions across different age groups31,34. A tangential to normal skull conductivity ratio of 10:1 was obtained from a three concentric sphere current-flow phantom and validated with in vivo data from monkey and human brains35. An anisotropy ratio of 3:1 has been utilized in previous studies7,36, but other reports proposed an even lower ratio of 1.6:137–39. Subsequent work suggested that assumed 10:1 ratio was likely overestimated, indicating a more realistic value of 1.8:140 which was also adopted for EEG source localization41. Furthermore, the constant skull conductivities across the three orthogonal directions have been considered up to frequencies of 10 kHz42. Uncertainties in skull conductivity studied on simplified human head phantom have been shown to directly affect spatial localization, leading to error of several centimeters, particularly in depth43. The recent paper44 concluded that the inaccuracies in skull conductivity can lead to significant errors in interpreting EEG data and skull anisotropy further complicates the interpretation of EEG signals. Moreover, the sensitivity of the forward solutions correlates with the changes in the inverse solution44. The study45 presented that an error in skull conductivity up to 20% is acceptable for FEM forward models with average discretization errors down to 3 mm. The paper10 concluded that the anisotropy of the skull resembles a deeper and weaker dipole, thus, challenging to accurately interpret the source in EEG source analysis. Moreover, the effect of anisotropy increases with the depth of the source and with the amount of anisotropic tissue surrounding the source. Variability in the spongiform bone affected both forward and inverse solutions, with neglect of this variability leading to substantial relative errors with discrepancies up to 85%46. Comparisons between isotropic and anisotropic realistic phantoms demonstrated that skull anisotropy exerts a stronger influence on radial than on tangential dipoles, and that increasing eccentricity amplifies the diffusion of the potential distribution47. Similarly, analyses of anisotropic and isotropic single-layer sphere phantom revealed that anisotropy introduces errors in both dipole position and moment48. The error 2–6 mm in localization was proven due to smoother patterns and a lower dipolar moment 9.5 nA.m instead of 10, which was caused by a decrease in the maximum of surface potentials. The role of skull inhomogeneities, particularly sutures, has been highlighted as critical for accurate geometry representation near electrodes45. Omitting sutures has been shown to induce substantial errors, reaching up to 67.7% in the forward model and 4.1 cm in source localization47. Similarly, exclusion of fontanelles has been reported to produce maximum source errors of up to 3.6 mm49, with the largest localization inaccuracies occurring directly beneath the fontanelles50. Consistently, incorporating fontanelles into source analysis has been found to enhance localization accuracy51. Precise forward modeling is a prerequisite for accurate EEG source localization, particularly when aiming to reconstruct functional brain networks in small animals such as rats especially in case of more translationally valid and less invasive EEG recording methods (above the skull implants). The specific anatomical features of the rat skull—its complex shape, varying thickness, and anisotropic conductivity—pose unique challenges not yet systematically addressed. While human EEG studies43–51 have long considered skull inhomogeneity and anisotropy as critical factors influencing source reconstruction accuracy, these aspects remain underexplored in rodent research. Given the increasing use of rats in translational neuroimaging and pharmaco-EEG, it becomes essential to quantify the effects of skull tissue properties on electric field propagation. This study aims to address the following key questions regarding forward modeling of the rat EEG: What is the effect of neglecting the skull in the forward model? What is the effect of neglecting the skull in the forward model? How do skull inhomogeneities and sutures influence the measured electric potential? How do skull inhomogeneities and sutures influence the measured electric potential? How does skull anisotropy affect the results compared to the isotropic phantom? How does skull anisotropy affect the results compared to the isotropic phantom? Which dipole positions and orientations are mostly influenced by mitigation of the head inhomogeneity and skull anisotropy? Which dipole positions and orientations are mostly influenced by mitigation of the head inhomogeneity and skull anisotropy? Is there any relation between position and direction of a testing dipole with respect to the highest leadfield differences? Is there any relation between position and direction of a testing dipole with respect to the highest leadfield differences? What is the impact of the skull thickness? What is the impact of the skull thickness? How sensitive are the simulations to various anisotropy ratios reported in the literature? How sensitive are the simulations to various anisotropy ratios reported in the literature? Methods Phantom geometries To evaluate the influence of inhomogeneity and anisotropy of skull tissue on forward solution, excluding the geometry effect, the numerical simulations were performed at first on five cuboid-shaped phantoms approximating a rat’s head (Fig. 1). This approach follows the well-known methodology widely used in human head representation4,8,48, where simple geometries, typically a sphere, are employed to isolate the effects of tissue properties from anatomical complexity. However, since the rat head geometry is far from a spherical shape, a cuboid representation was adopted as a more appropriate first-order approximation. Moreover, the geometric simplicity of the cuboid phantom provides a controlled environment in which the effects of material properties can be clearly assessed. In particular, transitions between isotropic and anisotropic conductivity can be directly attributed to changes in tissue properties rather than by complex geometry. The cuboid geometry also facilitates systematic investigation of dipole orientation effects, as the principal axes of the geometry are aligned with the dipole orientations, enabling clearer interpretation of directional dependencies. Cuboid-shaped and realistic phantoms differing in the skull material properties: Level 0 with isotropic skull whose electric conductivity is the same as the brain (a), Level 1 with isotropic skull and conventional value of electric conductivity (b), Level 2 with inhomogeneous skull (c), Level 3 with isotropic skull including the suture (d), Level 4 with anisotropic skull (e). Realistic skull phantom represents the isotropic case (f), and realistic skull phantom based on anisotropic conductivities σx_RedWall, σy_YellowWall, σz_BlueWall (g). All five phantoms are composed of the isotropic brain with the dimensions 24 mm × 10.1 mm × 14.48 mm, and the electric conductivity 0.33 S/m. For the skull surrounding the brain, the thickness of 0.72 mm is considered. These five phantoms differ in the skull material properties: Level 0 (Fig. 1a) – the skull has the same electric conductivity as the brain. Level 0 (Fig. 1a) – the skull has the same electric conductivity as the brain. Level 1 (Fig. 1b) – the skull is isotropic with the electric conductivity 0.0042 S/m52. Level 1 (Fig. 1b) – the skull is isotropic with the electric conductivity 0.0042 S/m52. Level 2 (Fig. 1c) – the skull is divided into three compartments. Each one is defined by its specific electric conductivity to mimic different skull thicknesses. Level 2 (Fig. 1c) – the skull is divided into three compartments. Each one is defined by its specific electric conductivity to mimic different skull thicknesses. Level 3 (Fig. 1d) – the skull is isotropic and considers the suture tissue with the electric conductivity 0.02 S/m31. Level 3 (Fig. 1d) – the skull is isotropic and considers the suture tissue with the electric conductivity 0.02 S/m31. Level 4 (Fig. 1e) – the skull is anisotropic. The cuboid-shaped phantom is defined by three electric conductivity tensors according to the orientation of the phantom’s walls. Level 4 (Fig. 1e) – the skull is anisotropic. The cuboid-shaped phantom is defined by three electric conductivity tensors according to the orientation of the phantom’s walls. The phantom Level 2 depicted on Fig. 1c was implemented as a single equivalent layer, which was further subdivided into three compartments to address regional variations of the skull53. In each compartment, it was assumed that the single layer is homogeneous and characterized by an effective electric conductivity. The effective properties of each compartment were determined by considering the relative contributions of the three skull tissues (inner compacta, spongiform, outer compacta). The thicknesses of these skull tissues were obtained from a single CT dataset of a rat skull. The dataset was divided into 41 slices, and in each slice, 20 thickness measurements per skull tissue were performed, resulting in 820 samples per skull tissue (2460 measurements overall). These measurements were averaged to determine representative thicknesses of the individual tissue layers for each compartment. The resulting thicknesses were proportionally rescaled so that their sum equaled a total skull thickness of 0.72 mm, consistent with the values used across all cuboid-shaped phantom Levels. Based on the proportionally rescaled values, the effective electric conductivity of each compartment was calculated as a thickness-weighted average of the electric conductivities of the individual skull tissues. As a result, each compartment is represented by a single value of effective electric conductivity, which is a combination of thickness and electric conductivity of three tissues. The resulting electric conductivities were σCompI = 0.0081 S/m, σCompII = 0.0078 S/m, σCompIII = 0.0097 S/m. A detailed description of the procedure is provided in the Supplementary Files. The phantom Level 3 (Fig. 1d) is isotropic as the phantom Level 1 with the electric conductivity 0.0042 S/m. Moreover, the phantom Level 3 considers the suture tissue with the electric conductivity31 0.02 S/m that was created by approximating the suture in CT scans using block shape. The phantom Level 4 (Fig. 1e) is anisotropic and respects the orientation of the anisotropy related to the phantom walls. Each wall is characterized by a diagonal electrical conductivity tensor defined in the global coordinate system. Specifically, the conductivity tensors are given by σx_RedWall = diag(0.0042, 0.042, 0.042) S/m, σy_YellowWall = diag(0. 042, 0.0042, 0.042) S/m, and σz_BlueWall = diag(0.042, 0.042, 0.0042) S/m, where the off-diagonal components are zero and the principal directions of anisotropy are aligned with the global coordinate axes. The conductivity tensors correspond to an anisotropy ratio of 10:1, with reduced conductivity in the normal direction to each wall and higher conductivity in the tangential directions. In order to preserve correct current flow, the edges and corners of the phantom Level 4 are subdivided. Each edge was split into two segments under a 45-degree angle, and the corners were divided into six segments to properly connect the edges. Further, we also investigate two conditions of the realistic skull phantom to facilitate a deeper study of skull anisotropy. The purpose of the realistic phantom is not to achieve a detailed anatomical description, but rather to act as a bridge between simple and realistic head. In addition, detailed anisotropic geometries of the rat skull have not been well established, unlike the human skull, which has been extensively studied, especially regarding the white matter7,11 and the skull41,47,48. Therefore, increasing the complexity of the geometry beyond the current level will imply the need to make more assumptions. Similar to cuboid-shaped phantoms, both conditions of realistic ones are based on the same geometry, differing in skull material properties, with an identical brain geometry assumed in both cases (electrical conductivity 0.33 S/m). The geometry is based on our own CT dataset, allowing full control of the segmentation and geometry construction pipeline. The geometry derived from the CT scan is converted into a voxel geometry with 1 mm edge length, offering a consistent and computationally efficient representation. Although more detailed rat head phantoms have been reported in the literature, the geometry adopted in this study was intentionally chosen to balance anatomical detail with controlled modification of material properties. Specifically, the use of a single consistent geometry enables systematic investigation of skull anisotropy without introducing unnecessary variations from geometry-specific differences. The realistic skull phantom depicted in Fig. 1f represents the isotropic case with electric conductivity 0.0042 S/m. In contrast, the phantom shown in Fig. 1g incorporates anisotropy defined by three electric conductivities σx_RedWall, σy_YellowWall, and σz_BlueWall respecting the anisotropy orientation regards to the geometry. Note that the front part of the skull colored light blue is isotropic since it is far from the excitation dipoles and sensing electrodes to decrease computational demands. The black voxels represent the transition between voxels with different anisotropy tensors (σx_RedWall, σy_YellowWall, and σz_BlueWall) to preserve the correct current flow. Here, the black voxels are defined by an isotropic tensor σ = (0.0042 I) S/m, where I is a 3 × 3 identity matrix, ensuring equal current flow in all directions. Sensing electrodes and modelling of neural activity of the brain The cuboid-shaped phantoms are equipped with 34 electrodes to measure electric potential and 2 ground electrodes (GND) placed on the surface of the skull (Fig. 2a). This electrode layout corresponds to the one used in our laboratory for EEG recordings in living rats, ensuring compatibility between the simulation and experimental setups. To measure electric potential in various heights, we implement two rows of three electrodes on both sides of the cuboid-shaped phantoms, i.e., 12 side electrodes (Fig. 2b). The rest of the measuring electrodes are placed on the top of the geometry including two ground electrodes. The EEG is measured as voltages at the measuring electrodes relative to the conductively connected ground electrodes (GND1 and GND2). Sensing electrode layout and exciting dipole positions: top view of the electrodes for the cuboid-shaped phantom (a), perspective view of the electrodes for the cuboid-shaped phantom (b), top view of the exciting dipoles for cuboid-shaped phantom (c), side view of the exciting dipoles for cuboid-shaped phantom (d), electrode layout of sensing electrodes for realistic phantom (e), and dipole positions for realistic phantom (f). The neural activity is modeled by 8 excitation dipoles located within the brain, with their placement chosen to reflect realistically expected sources of activity in the living brain, specifically targeting key nodes of the rat default mode network (DMN) and primary sensorimotor generators. Dipole D6, located centrally in the anterior portion of the geometry and x-oriented, corresponds to the prelimbic cortex (Cg1/Cg2 region), which serves as an anterior DMN hub involved in cognitive control and internal state processing14. Dipole D2, positioned centrally in the posterior region and z-oriented, represents the retrosplenial cortex (RSD/RSG), a posterior DMN hub critical for spatial navigation and memory integration that exhibits the highest degree of connectivity within the rat DMN14,20. The symmetrical dipole pairs D4 and D8, oriented along the y-axis and positioned laterally close to the brain-skull interface, represent the left and right auditory/temporal association cortex, regions involved in auditory processing and multimodal sensory integration that also participate in DMN activity12. Dipoles D5 and D7, which are also y-oriented and symmetrical, represent the left and right primary somatosensory/motor cortex, respectively. These areas generate characteristic oscillations during locomotion and serve as the dominant generators of movement-related cortical activity7. The deeper dipoles D1 and D3 correspond to the auditory thalamus (medial geniculate nucleus). Given the uncertainty regarding thalamic dipole orientation and the limited scalp EEG sensitivity to deep subcortical sources21, these dipoles are positioned with different orientations: D1 is placed vertically (perpendicular to the skull plane) and D3 horizontally (parallel to the skull plane) to test the effect of source orientation on surface potential distributions and localization accuracy. Figure 2c shows the location of the dipoles from the top view of the cuboid-shaped phantom. The dipole pairs D4-D8 and D5-D7 are y-oriented and symmetrical with respect to the geometry. Moreover, the dipole pair D4-D8 is very close to the brain-skull interface. The positions of the dipoles D1-D3 are symmetrical with respect to the geometry, but the dipoles are oriented along different axes. Finally, the dipoles D2 and D6 are located in the center of symmetry of the cuboid-shaped phantom. Figure 2b illustrates the varying depths of individual dipoles, enabling the examination of the impact of source depths on electric potentials in relation to inhomogeneity and anisotropy. To investigate the inhomogeneity and anisotropy of the rat head in more detail, a volumetric grid of infinitesimally small dipoles is defined throughout the head model. Lead field values were computed for each dipole using the reciprocity theorem, see the section below. Only dipoles positioned below the skull were retained as the test set, ensuring that the analysis was confined to intracranial sources. This dipole configuration enables systematic evaluation of source localization accuracy across functionally relevant brain regions with varying depths and orientations. Furthermore, the phantom’s electrode layout mirrors the actual electrode arrangement used in our laboratory’s in vivo recordings, ensuring that the spatial sampling of signals in the phantom reflects real experimental conditions. By anchoring the phantom’s design to these biologically relevant structures, we capture important features of the rat’s brain functional architecture – from DMN midline hubs to active sensorimotor regions – thereby making the phantom’s EEG outputs more representative of true neural activity. The realistic skull phantoms exploit the same electrode layout as cuboid-shaped phantoms, i.e. 34 measuring electrodes and 2 GND electrodes touching the skull surface as shown in Fig. 2e. Furthermore, dipole positions in the realistic skull phantoms are identical to those in the cuboid-shape phantoms, as illustrated in Fig. 2f. Full-wave simulations and reciprocity approach The computation of forward problem is performed in the CST Studio Suite® Low Frequency (LF) Frequency Domain Solver54 which solves the full-wave Maxwell’s equations by the finite element method (FEM) in the frequency domain. This solver was used for all simulation-based analyses presented in this paper, which include both cuboid and realistic head phantoms, as well as for the evaluation of skull thickness and anisotropy ratio effects. Subsequently, the solver was experimentally validated using EEG measurements and was cross-compared with other forward solvers, as will be shown in the following section. The geometry is discretized by a tetrahedral mesh and an adaptive mesh refinement is exploited to ensure accurate results. The adaptive mesh works iteratively. Two consecutive simulations are performed and their 3D field distributions are compared. When differences in fields are detected, the mesh is refined in regions with high error. This process is repeated until two consecutive simulations produce identical results, thus the adaptive mesh ensures sufficiently accurate mesh. The adaptive mesh has been consistently adopted for all simulations carried out in the CST Studio Suite®54. To solve full-wave Maxwell’s equations, the frequency, permittivity, and permeability have to be defined. Those parameters are set similarly to those in our previous studies55,56 related to the EEG forward model of rat head. We set the frequency to 1 kHz. Since the biological tissues are nonmagnetic materials and dissipation current can be neglected, relative permeability and permittivity are set to 1 for all tissues, respectively. Thus, only the electric conductivity of tissues is considered. The dipole is represented by two cylinders with radii of 0.22 mm and 0.051 mm separated by a gap of 0.1 mm, with a total length of 1.3 mm. Each dipole is excited gradually with the current of 1 µA, and the resulting electric potential is measured by ideal-point electrodes. As mentioned above, a volumetric grid of testing dipoles was considered to investigate the inhomogeneity and anisotropy of the rat head. To this end, the reciprocity theorem was applied to make the computations feasible. The electric field intensity resulting from current excitation of a pair of electrodes was computed within the brain volume. This was done for all electrodes paired with the GND1 electrode. The corresponding lead field can be easily derived from the computed electric field intensity by appropriate scaling4. Here, the advantage of the full-wave formulation of the forward problem can be fully utilized, since the electric field intensity is directly accessible in the solution and no numerical differentiation (needed in case of the quasi-static form) of the electric potential is required for the implementation of the reciprocity theorem. Validation of full-wave simulations using EEG measurement on physical phantom and EM software To experimentally support our work, the cuboid-shaped phantom Level 0 (Fig. 1a) was fabricated (Fig. 3). Since the goal was to fabricate the phantom based on conventionally available techniques assuming a linear behavior of the phantom56, the size of the phantom was scaled by a factor 5. Physical phantom fabrication and EEG measurement: top view on box of shell with electrodes and dipoles (a), bottom view on box of shell with electrodes and dipoles (b), lid of shell with excitation dipoles (c), bottom view on complete phantom before filling by agar-sodium chloride mixture (d), sensing electrode (e), excitation dipole (f), EEG measurement setup and testing of physical phantom (g). The phantom is composed of a non-electrically conductive shell, sensing electrodes (Fig. 3e), excitation dipoles (Fig. 3f), and an agar-sodium chloride (NaCl) mixture for mimicking the brain and skull properties. Let’s remember that both these tissues have the same electric conductivity for this phantom. The fabrication procedure can be summarized as follows: The shell consists of a box and a bottom lid. Both parts were printed with a 3D printer Original Prusa i3 MK3S based on the fused filament fabrication (FFF) technology and the polyethylene terephthalate glycol (PETG) filament. The shell consists of a box and a bottom lid. Both parts were printed with a 3D printer Original Prusa i3 MK3S based on the fused filament fabrication (FFF) technology and the polyethylene terephthalate glycol (PETG) filament. 36 holes were drilled in the box of the shell for electrodes. Electrodes are based on copper rivets of the diameter 3 mm. The lower part of electrodes is platinum plated since cut rivet part touches an agar-sodium chloride mixture. The top part of electrodes is equipped by conventional pins which are soldered to rivets. The electrodes were fixed to the box of the shell by a hot melt glue gun. 36 holes were drilled in the box of the shell for electrodes. Electrodes are based on copper rivets of the diameter 3 mm. The lower part of electrodes is platinum plated since cut rivet part touches an agar-sodium chloride mixture. The top part of electrodes is equipped by conventional pins which are soldered to rivets. The electrodes were fixed to the box of the shell by a hot melt glue gun. 6 holes and 2 holes were drilled in the printed box and the bottom lid of the shell, respectively. These holes are used for excitation electric dipoles which represents the neural activity of the brain. The dipoles whose surface was platinum plated are based on a semirigid coaxial cable. The first arm of the dipole is created by an inner conductor of the coaxial cable with the diameter 0.51 mm. The second arm of the dipole is created by the outer conductor of the coaxial cable with the diameter 2.2 mm. The length of each dipole arm is 3 mm. The rest of the coaxial cable is isolated from the surrounding environment. The excitation dipoles to desired positions were fixed to the box and the lid of the shell by a hot melt glue gun. 6 holes and 2 holes were drilled in the printed box and the bottom lid of the shell, respectively. These holes are used for excitation electric dipoles which represents the neural activity of the brain. The dipoles whose surface was platinum plated are based on a semirigid coaxial cable. The first arm of the dipole is created by an inner conductor of the coaxial cable with the diameter 0.51 mm. The second arm of the dipole is created by the outer conductor of the coaxial cable with the diameter 2.2 mm. The length of each dipole arm is 3 mm. The rest of the coaxial cable is isolated from the surrounding environment. The excitation dipoles to desired positions were fixed to the box and the lid of the shell by a hot melt glue gun. The box of the shell was closed by a lid with the help of a hot melt glue gun. The box of the shell was closed by a lid with the help of a hot melt glue gun. The agar-sodium chloride mixture mimicking gray matter was prepared carefully according to the procedure described in our previous study56 and poured into the shell. After 18 h, when the phantom was cooled to room temperature, it was measured. The agar-sodium chloride mixture mimicking gray matter was prepared carefully according to the procedure described in our previous study56 and poured into the shell. After 18 h, when the phantom was cooled to room temperature, it was measured. Each excitation dipole was sequentially excited using the generator IQ SIGLENT SFD60022X producing a 1 kHz AC sine wave with peak amplitude of 40 mV. The frequency of 1 kHz was selected to eliminate an oxidation-reduction reaction affecting the electrode impedance. Moreover, exciting voltage of 40 mV was chosen to ensure sufficient signal-to-noise ratio and a linear regime of physical phantom. The dipoles produced the electric potential measured by BioSDA09, a 32-channel digital EEG amplifier (MI Ltd., Prague, Czech Republic) operating at a sampling frequency of 5 kHz and connected via a cable (Data Sciences International, St. Paul, Minnesota, United States). The EEG measurement setup and testing of physical phantom Level 0 is shown in Fig. 3g. EEG was recorded at 34 electrodes as time-dependent voltage signals referenced to two electrically connected ground electrodes. Similar to the simulation in chapter II.B, the odd-numbered electrodes were referenced to GND1, and even-numbered electrodes to GND2. EEG recording took 60 s for each dipole, however, only 30 s from the center of time interval were chosen for post-processing due to the signal stabilization. Fourier transform was further applied to the time intervals and the absolute value of 1 kHz frequency component was taken as the amplitude measured on electrode. To acquire the precise shape of the physical phantom and the exact positions of the excitation dipoles and electrodes, the phantoms were scanned one day after EEG measurement using the CT Siemens Somatom Definition Flash with the help of the CT department at the Institute for Clinical and Experimental Medicine in Prague. Native Q2 3D images of physical phantom were acquired with acquisition matrix 654 × 1024 × 1024. Tissues were segmented in 3D Slicer software57 into three compartments: a brain, dipoles, and electrodes, which were subsequently used for simulations. Besides EEG measurement, the full-wave simulations in CST Studio Suite are further verified by FieldTrip toolbox52 implemented in MATLAB58, COMSOL Multiphysics®59, and the Electro-QuasiStatic (EQS) solver in CST Studio Suite®54. All software packages are based on the FEM numerical technique except the Fieldtrip toolbox. However, we incorporated the SimBio toolbox60 with Fieldtrip to perform FEM simulations. The FieldTrip toolbox and COMSOL Multiphysics solve the Poisson’s equation, thus the frequency, relative permittivity, and relative permeability are neglected. In addition, the neural activity is represented as analytical point dipoles and the electric potential is measured by ideal-point electrodes. The EQS solver in CST Studio Suite ignores the time derivative of the magnetic field in the Faraday law, hence the computed electric field is curl-free in the whole space54. Moreover, the finite dimensions of the dipoles are allowed, thus, the dipoles were represented in 3D corresponding to the sources in EEG measurement. Similar to the Fieldtrip toolbox and COMSOL Multiphysics, the electrodes were implemented as ideal points. Two standard criteria61 in forward modelling were employed to assess the difference between EEG measurement and simulations. The Relative Difference Measure (RDM) is the first criterion evaluating the difference in shape between two data sets: where φnum is computed electric potential, φmeas corresponds to measured electric potential, and N denotes the total number of measuring electrodes. The overall RDM is obtained as the average of the RDM computed for all electrodes. The second criterion is the magnification factor (MAG) evaluating the error in magnitude: Note that, when the simulation results equal the measured ones, the RDM = 0 and MAG = 0. Leadfield comparisons on a volumetric grid A grid containing 3600 testing dipoles was considered to compare simulated leadfield across conditions: Level 0, Level 1, and Level 4. Within this setting each position inside the brain was represented by a local leadfield matrix, where s the number of electrodes and nOri=3 is the number of dipole orthogonal orientations. The spatially local comparisons were provided via comparison of the local lead field matrices. First, Normalized Euclidean distance between concatenated local matrices was computed. For each i-th dipole position and j-th phantom Level, the leadfield gives a 102-element vector (3 orientations × 34 electrodes) and the normalized Euclidean distance can be expressed: Both vectors are first normalized to unit length. Thus, amplitude differences are discarded, and only the shape (the topographical pattern) of the scalp potential distribution is compared. The resulting spatial map is then plotted as contours. Projection onto the three orthogonal slices is computed using averaging for data aggregation across the hidden dimension. Second, we searched for the worst-case orientation of each dipole, the orientation that maximizes . This corresponds to the case in which using an incorrect head model leads to the largest error. The result is visualized as a vector field in three orthogonal slices. For subsequent analyses, the dipole orientations are constrained to point outward from the phantom. This has no effect on the physical interpretation of the orientation, as the sources are assumed to be oscillatory on the one hand, and it enables us to quantify the relationship between position and orientation via correlation analysis on the other hand. Third, the relationship between the worst-case orientation and dipole position is depicted in scatter plots and quantitatively evaluated using the Pearson correlation coefficient. Results This section presents a verification of the full-wave simulation in CST Studio Suite using EEG measurement and electric potentials obtained from other software. Further, the comparison of results from full-wave simulations in CST Studio Suite for the cuboid-shaped phantom and the realistic skull one. The cuboid-shaped phantom is defined in five Levels differing in skull material properties. The realistic skull rat head phantom is investigated from the viewpoint of the anisotropy effects. Finally, the relationship of skull thickness and anisotropy ratio is investigated for the cuboid-shaped phantom and relation between dipole position and the worst-case orientation is investigated on the whole brain volumetric grid. The computed and measured absolute electric potentials at individual electrodes are provided in the Supplementary files. These values served as the basis for the statistical representation presented here. In cases where additional clarity is required for data interpretation, the absolute electric potentials are also presented here. Verification of full-wave simulation by a physical experiment To verify our computation of forward problem performed in the CST Studio Suite® LF Frequency Domain Solver54, we compare the electric potentials with EEG measurement and simulations using other software on the cuboid-shaped phantom Level 0. The resulting electric potentials are normalized to the maximum amplitudes because individual software packages have various definitions of excitation. The obtained results are depicted in a boxplot representation in Fig. 4. The comparison clearly demonstrates that the simulations performed in all software is consistent with EEG measurement across all dipoles (D1 to D8) and electrodes showing only minimal differences, with the highest potentials at D3 and the lowest at D4. The interquartile ranges show that CST Studio Suite Electro-QuasiStatic (EQS), CST Studio Suite LF Frequency Domain Solver (FW), and COMSOL Multiphysics (COMSOL) solvers provide consistent results. In contrast, the results from FieldTrip (FT) show slightly more variability for some dipoles. The measured electric potentials on the physical phantom fall within the range suggested by the solvers, confirming the simulations reliability. Verification of full-wave simulation in CST studio suite by EEG measurement on physical phantom and simulations in different computational solvers for all dipoles. The agreement between the simulated electric potentials and measured ones across all dipoles and electrodes were assessed by RDM and MAG as shown in Table 1. The best matching results within RDM were obtained by EQS solver, followed by FW solver achieving almost identical results. The overall worst match was achieved by the FT solver, however, the difference from EQS is only 0.0034 (Table 1). According to the MAG criterium, the FW solver showed the best agreement with the EEG measurement and the FT solver performed the worst showing nearly twice MAG value of the FW solver. Table 1Comparison of measured and simulated electric potentials.DipoleEQSFWFTCOMSOLRDMMAGRDMMAGRDMMAGRDMMAGD10.01980.03510.02000.03140.02620.05920.02070.0142D20.01460.07000.01470.07020.01390.10980.01500.0834D30.01280.04920.01280.04930.01520.10760.01610.0416D40.00350.02060.00360.02120.01100.06660.00680.0512D50.00550.01090.00540.01070.01430.04580.00710.0236D60.00920.00600.00920.00510.00840.00680.00920.0007D70.01170.00220.01180.00370.01340.00580.01470.0062D80.00390.01310.00400.01360.00530.00670.00550.0165 AVG 0.0101 0.0259 0.0102 0.0256 0.0135 0.0510 0.0119 0.0297 Comparison of measured and simulated electric potentials. Cuboid-shaped and realistic phantoms The results for five levels of cuboid-shaped phantoms are presented in Fig. 5a. In forward modeling, the skull is typically assumed to be isotropic and homogeneous, thus we consider the phantom Level 1 as the reference one. The largest difference in simulated electric potentials is exhibited by the phantoms Level 0 and Level 4, hence neglecting of the skull properties and anisotropy may impact the forward model accuracy. On the other hand, the most consistent results with the phantom Level 1 are evident for the phantoms Level 2 and Level 3. The interquartile range is the smallest at dipoles D4 and D8 located closest to the skull surface. Their shallow location causes the generated electric potentials to be highly localized, mainly affecting a single electrode. On the other hand, deeper dipoles such as the D1 or D3 produce more distributed fields that impact multiple electrodes leading to larger interquartile range. Simulation of phantoms: box plot presenting electric potentials simulated across all dipoles and Levels for cuboid-shaped phantom (a), and realistic phantom (b). Detailed bar graph showing absolute values of electric potential on individual electrodes for dipole: D1 - even electrodes (c), D2 - even electrodes (d), D4 - even electrodes (e), D8 - odd electrodes (f), D1 - odd electrodes (g), D3 - odd electrodes (h). Note that, detail graphs for all dipoles and electrodes for cuboid-shaped phantoms and realistic ones can be found in supplementary files. Neglecting the skull layer (Level 0) has the strongest effect for the deep dipoles (D1, D3) that are distant from the electrodes. As the distance of the dipole to the skull decrease, the effect becomes weaker as observed for dipoles closer to the skull (D2, D5-D7). This behavior can be explained by the fact that shallow dipoles mainly affect nearby electrodes, whereas their influence on distant electrodes remains negligible. When comparing inhomogeneous (Level 2) and homogeneous (Level 1) phantoms, the resulting electric potentials are statistically very similar across all electrodes. The similarity can be attributed to the relatively small contrast in electric conductivities among skull regions which are approximately twice as high as the homogeneous phantom Level 1. Obviously, Level 1 and Level 3 (Fig. 5a) exhibit comparable electric potentials. However, when examining the detail distribution of electric field across electrodes, the differences are apparent on electrodes close to the suture tissue. The highest difference is observed at electrodes V4 and V6 for the dipole D1 (Fig. 5c). A similar effect is evident for dipole D2 and the electrodes P2 and R2 (Fig. 5d). The differences are primarily confirmed to electrodes close to the excitation dipole. In contrast, electric potential on electrodes situated on the suture tissue and far away from the dipole show only negligible influence of the suture. This can be seen in Fig. 5d for dipole D2, where electrodes F4, C2, V4, and V6 are located on the suture tissue, but the effect remains negligible. The strongest influence of anisotropy is observed when the dipole is located very close to an electrode. The evidence for this behavior is obvious for dipole D4 and electrode A3, and the dipole D8 and the electrode A4 as shown in Fig. 5d. On the other hand, with increasing distance between the dipole and the sensing electrode, the effect of anisotropy decreases. A minor effect of anisotropy is obvious from Fig. 5c for dipole D1 located deep within the brain and for the odd electrodes positioned far from the dipole. By comparing phantoms Level 1 and Level 4, it is obvious that the anisotropy does not affect all electrodes in the same way, i.e., the electric potential at certain electrodes is lower due to anisotropy, but other electrodes exhibit higher electric potential. For instance, from the simulation of the dipole D1 (Fig. 5c) it is obvious that the electric potential at the side electrodes (P5, P7, A1, A3, T1, T3) is higher compared with the phantom Level 1 while the electric potential at the other electrodes is lower. The resulting comparison of electric potentials for realistic phantoms is shown in Fig. 5b. The isotropic realistic phantom consistently produces higher electric potentials across all dipoles except dipole D1, in particular for dipoles D2 and D3. When we compare the realistic phantom with cuboid-shaped one, the effect of anisotropy is even more pronounced, indicating that the influence of anisotropy increases with the complexity of the geometry. The interquartile ranges are smallest for dipole D4 and D8 and largest for D1 and D3, reflecting the influence of dipole depth within the brain similar to the cuboid-shaped simulations. Considerable differences in the medians can be observed for symmetrical pairs of dipoles D5, D7 and D4, D8 pointing towards the side electrodes and normal to the top electrodes. Dipole D2 located closest to the top electrodes and directed toward them produces a large difference in the interquartile range while the medians are similar for both phantoms. Comparable medians are also observed for dipole D1, which similarly points directly toward the top electrodes. Note that the differences in the interquartile range are again attributed to the differing depths of the dipole D1 and D2. Similar medians are also observed for dipole D6 in both isotropic and anisotropic phantoms, but in this case, it is due to the dipole not pointing toward the electrodes and affecting only a few frontal electrodes. Skull thickness and anisotropy ratio effect We further analyze the effect of the skull thickness on the computed electric potentials for all levels of cuboid-shaped phantoms. The skull thickness is varied towards the brain geometry for three values – 0.72 mm (original), 1.44 mm, and 2.16 mm. The inward expansion of the skull was selected to maintain the same positions of the electrodes, ensuring comparability between all thickness conditions. Since the brain volume decreased as the skull volume increased, the original dipoles D4, D5, D7, and D8 are situated in the skull geometry, thus only the dipoles D1, D2, D3, and D6 are considered for this type of investigation. Note that the phantom Level 0 is not considered since the skull and brain geometries are defined by the same electric conductivity. Thus, the skull thickness has no impact on electric potentials in that case. As the skull thickness at the isotropic phantom Level 1 increases (Fig. 6a), the electric potential at all electrodes increase. Level 2 exhibits a similar behavior to Level 1 as shown in Fig. 6b. However, upon closer inspection, a reduced interquartile range can be observed for dipoles D2 and D6 as the skull thickness increases. The reduction of electric potential is caused by higher electric conductivity compared to the isotropic phantom Level 1. In contrast, the effect is not evident for dipoles D1 and D3, as they are located deeper in the brain. A more noticeable change appears in the phantom Level 3 with sutures. Dipoles D2 and D6 are positioned at the center of the cuboid phantom, matching the location of the suture. In case of a skull thickness of 2.16 mm, those dipoles are very close to the suture leading to a significant rise in the interquartile range for both dipoles. In contrast, electric potentials for dipoles D1 and D3 are similar to the isotropic phantom Level 1, as they are located farther from the sutures. Parametric simulation of cuboid-shaped phantom skull thickness for: Level 1 (a), Level 2 (b), Level 3 (c), Level 4 (d), and anisotropy ratio of Level 4 (e). Irregular behavior is seen when increasing the anisotropic skull thickness at the phantom Level 4 (Fig. 6d). As the thickness increases, the median of electric potential decreases for dipole D1 while it increases for dipoles D2 and D3. However, for dipole D6, electric potentials exhibit alternating behavior. Dipole D1 points toward the upper electrodes where the surface has higher conductivity along the z-direction. On the other hand, dipole D3 points away from both the top and side electrodes. Therefore, it’s less affected, leading to a nearly linear increase in median potential similar to isotropic phantom Level 1. Dipoles D2 and D6 are close to the skull when its thickness is 2.16 mm. This closeness may explain the greater rise in electric potential. These findings show unexpected behavior at the interface with anisotropic tissue. Notably, dipoles are affected regardless of their orientation relative to the electrodes. Dipole D6 is turned away from them, while D2 faces the top electrodes. The position-orientation relationship is elaborated in the next section in more detail. In the following, we analyze the impact of anisotropy ratio of the skull, as different anisotropy values have been reported in the literature. Specifically, the following anisotropy have been considered: 1:1 (isotropic), 1.6:137–39, 1.8:140,41, 3:17, 36, 10:135. Note that the default electric conductivity is 0.0042 S/m, which is scaled according to the specific anisotropy ratio. The comparison of electric potential for different anisotropy ratios in the cuboid-shaped phantom is shown in Fig. 6e. The ratios 1.6:1, 1.8:1, and 3:1 are close to the isotropic phantom Level 1. However, a clear difference appears for the 10:1 ratio. When looking at individual dipoles, the median goes up for dipoles D1, D2, and D6, while it goes down for D3. The interquartile range increases for D2, but it decreases for D1, D3, and D6. Notably, dipoles D1 and D3 are at the same depth, and both show a decrease in interquartile range. However, the median increases for D1 and decreases for D3. This behavior can be explained by their orientation. D1 points toward the electrodes, while D3 points away. Position vs. orientation leadfield comparison The effects of rat head inhomogeneity (Level 0 vs. Level 1) and anisotropy (Level 0 vs. Level 4 and Level 1 vs. Level 4) were further investigated on a volumetric grid containing 3600 testing dipoles. Three questions were examined: the spatial distribution of the leadfield differences, the worst-case dipole orientations and their spatial distributions, and the relationship between dipole position and orientation within the brain. In Fig. 7, the spatial distribution of the normalized Euclidean distance of the leadfields can be observed. There is a clearly observable increase in the differences for dipoles closer to the inner surface of the skull when inhomogeneity is neglected (see Fig. 7b). The differences are even more pronounced for dipoles close to the electrodes when both inhomogeneity and anisotropy are neglected (see Fig. 7b). Here, two factors are intertwined: the distance from the inner skull surface and the distance from the sensing electrodes. Finally, when anisotropy is specifically neglected, the leadfield difference appears to decay monotonically with the distance between dipoles and electrodes. Effects of neglecting rat head inhomogeneity (a), combined inhomogeneity and anisotropy (b), and anisotropy alone (c). Local lead field differences are depicted using contour plots. The differences increase toward the skull boundaries (a) and in the vicinity of the electrodes (b,c). The orientations corresponding to maximal differences are indicated by arrows. The worst-case orientations are predominantly perpendicular to the skull inner surface (radial dipoles). However, this pattern changes near the electrodes, where the orientations appear more random and can even become parallel (tangential dipoles) to the inner skull surface when anisotropy is neglected (c). Scatter plots illustrate the relationship between dipole positions and orientations projected onto the corresponding axes. A clear relationship between dipole direction and position is observed, further supporting that the most affected dipoles are predominantly oriented perpendicular to the inner skull surface (radial dipoles). There is a clearly observable structure in the worst-case dipole orientations within the brain volume. In general, dipoles that are perpendicular to the inner surface of the skull (radial dipoles) are most affected by neglecting the skull’s electrical properties, see the quiver plots in Fig. 7. At the same time, there is a rather random distribution of dipole orientations in the vicinity of the electrodes. Interestingly, when skull anisotropy is neglected, tangential dipoles (parallel to the inner skull surface) are most affected ones, see Fig. 7c. To investigate the relationship between dipole position and its most sensitive orientation, scatter plots in Fig. 7 were generated. Across all three comparisons, a clear relationship between position and projected orientation is visible, most prominently in the X and Y directions. Table 2 summarizes the quantification of this position–orientation relationship in the form of Pearson correlation coefficients. The main diagonals of all correlation matrices show much higher coefficients compared to the off-diagonal entries. This indicates that the relationship observed in the scatter plots is specific to corresponding position–orientation dimensions, rather than reflecting a general increase in differences across all dimensions. This is in good agreement with the quiver plots in Fig. 7. Table 2Pearson correlation between dipole positions and worst-case orientation.Level 0 vs. Level 1Level 0 vs. Level 4Level 1 vs. Level 4Dir xDir yDir zDir xDir yDir zDir xDir yDir zPos x+ 0.88+ 0.01+ 0.10+ 0.87+ 0.01+ 0.05+ 0.89+ 0.02+ 0.06Pos y− 0.01+ 0.89+ 0.02+ 0.01+ 0.88− 0.03+ 0.01+ 0.86− 0.03Pos z− 0.02− 0.01+ 0.75+ 0.01+ 0.00+ 0.80+ 0.02+ 0.01+ 0.77 Pearson correlation between dipole positions and worst-case orientation. Discussion Accurate forward modeling is essential for precise source localization and for mapping connectivity patterns in the rat brain. In the present study, we address several key questions posed in the introduction section regarding skull inhomogeneity, anisotropy, thickness, and their impact on measured electric potentials. The following discussion interprets our findings in the context of these specific questions. To answer the first question, we compared the cuboid-shaped phantom without the skull (Level 0) to the other levels consisting the skull (Fig. 5a). Neglecting the skull layer leads to notable differences of the electric potentials, particularly for dipoles located deeper in the brain (prelimbic cortex and thalamus in our case). However, the dipoles located closer to the skull, the difference between phantom Levels with and without skull becomes less significant, since the nearby electrodes are mainly affected. Overall, these findings highlight the critical role of the skull geometry in forward model. To address the second question, we compared inhomogeneous skull phantoms (Level 2 and Level 3) with the homogeneous one (Level 1) as shown in Fig. 5a. The results showed that the overall electric potential remained statistically similar across all electrodes indicating that the small conductivity variations across skull have negligible effect. However, local differences were observed at electrodes located near the suture tissues (somatosensory and motor cortices, retrosplenial dysgranular and visual cortices), particularly for dipoles positioned close to the skull. In contrast, electrodes located on sutures tissue, but farther from the dipoles, showed negligible changes, suggesting that the suture effect is spatially limited. Previous study46 investigated skull heterogeneity using a highly realistic human head with the skull represented by compact and spongiform bone. The spongiform bone (80% of the skull) region had spatially varying electric conductivity (0.004–0.015 S/m) while the compact bone had a constant electric conductivity of 0.003787 S/m. Comparison of homogeneous skull forward models resulted in relative errors up to 85% (0.01 S/m) and 57% (0.0055 S/m). In our case, however, such significant discrepancies were not observed. This could be due to the complexity, as the referenced study employs a highly detailed human head geometry including suture tissues, whereas our analysis is based on simplified rat head geometries, where individual effects were intentionally isolated (i.e. phantoms including only suture tissues or only skull heterogeneity). For this reason, a direct comparison with a simplified cuboid phantom is not appropriate. Our results are more consistent with the findings reported in the study focused on the influence of skull-conductivity misspecification45, where variations in human skull conductivity resulted in errors of approximately 20%. In addition, the study recommended to consider local variations in electric conductivity such as sutures in the regions with a high concentration of electrodes, which is also supported by our observations. Moreover, our study shows that the influence of sutures is the most evident when the dipole is located close to suture tissue and corresponding electrode. The effect of anisotropy on the forward model results depends primarily on the relationship between dipole and the electrodes. The strongest influence was observed when the dipole was located very close to an electrode, particularly for the limiting anisotropy ratios of 10:1, while substantially smaller effects were observed for more realistic anisotropy values (1.3:1, 1.8:1) as shown in Fig. 6e. This situation corresponds to the cortical sources which are regions of interest of preclinical research including functionally and anatomically distinct regions like the prefrontal cortex (executive functions, working memory), parietal cortex (spatial navigation, multimodal integration), visual cortex, Posterior Parietal (association, navigation), and temporal cortex. Deep dipoles showed only a mirror effect, particularly at distant electrodes. Furthermore, the analysis of anisotropy in relation to orientation of the dipole show that dipoles in the vicinity of the electrodes are mostly affected if their main direction is in parallel to the skull inner surface (tangential dipoles). Interestingly, the bigger distance from the electrodes the more perpendicular orientation to the skull of the most affected dipoles were observed. A comparison between the isotropic and anisotropic phantoms shown on Fig. 5a revealed that anisotropy does not affect all electrodes equally—while the electric potential decreases at some electrodes due to anisotropy, others exhibit an increase. Our results can be compared to previous human head studies62, in which the skull was represented by isotropic layers of spongiform and compacta. Specific values of isotropic conductivity assigned to these layers result in anisotropy ratio of 3:1, showing a weak effect in EEG source analysis. In contrast, we defined the rat skull as a single structure with anisotropy up to 10:1, producing stronger effects for dipoles close to electrodes, while lower anisotropy ratios resulted in correspondingly smaller effects. Species differences and geometry simplifications further contribute to the observed differences. However, both studies highlight that skull layer representation or anisotropy can influence forward model results. In contrast to the findings reported for the inverse solution10, our forward model results show the opposite trend – the highest differences in electric potentials were observed for dipoles located close to the electrodes, whereas only mirror effect of anisotropy was found for the deeper dipoles. Unlike the results in the study47 indicating a stronger impact of skull anisotropy on radial dipoles than tangential ones, our results showed comparable influence of anisotropy for both dipole orientations. Based on the volumetric investigation, the tangential dipoles are most prominent in the vicinity of the electrodes, while radial dipoles are the most affected ones in distant locations. This suggests that, the influence of anisotropy on dipole orientation in forward model is more complex than it was assumed. Our simulations with realistic phantom demonstrated that the influence of anisotropy is higher in realistic phantom compared to simplified phantoms such as cuboid-shaped phantom. Therefore, we support the findings of papers6–9 emphasizing the importance of exploiting realistic geometries to improve accuracy of the forward model. As the skull thickness increased in the inhomogeneous phantom Level 2 and Level 3 (Fig. 6b and c), higher differences in electric potentials were observed compared to the homogeneous phantom Level 1 (Fig. 6a) for dipoles located close to the skull surface. This effect was not evident for deeper dipoles placed farther from the skull. Moreover, significant increase in electric potential was observed for the dipole D2 located very close to suture tissue as a result of increased skull thickness. This could explain why the study47 reported forward model errors up to 67.7% if the suture tissue is neglected. When the skull thickness increased to 2.16 mm, bringing the dipoles very close to the surface, an increase in both the median and interquartile range of the electric potential was observed. This behavior matches the findings47, where higher eccentricity resulted in amplified diffusion of potential distribution. The accurate forward modeling achieved in this study carries substantial implications for preclinical EEG source localization, particularly when targeting functionally defined brain structures such as the rat default mode network (DMN) and sensorimotor hubs. By systematically evaluating the effects of skull inhomogeneity, anisotropy, and thickness across biologically motivated dipole locations, our results demonstrate how each of these properties differentially affects source reconstruction, depending on the spatial relationship between the dipole, skull, and electrode. For superficial cortical dipoles, such as those representing the auditory and temporal cortices (D4/D8), retrosplenial cortex (D2), and somatomotor cortices (D5/D7), skull anisotropy and local thickness variations, especially in regions overlying sutures, significantly influenced surface potentials. These findings underscore the importance of detailed skull representation when attempting to precisely localize cortical oscillatory generators, such as sensorimotor rhythms. In contrast, deep dipoles, such as those representing the thalamus (D1/D3) and prelimbic cortex (D6), were less sensitive to local skull inhomogeneities or anisotropies but were more affected by the presence or absence of the skull layer itself, reflecting the depth-dependent attenuation of scalp potentials. Notably, orientation-related differences for the thalamic dipoles were limited, suggesting that deep generators project diffusely and are inherently harder to localize accurately, regardless of modeling precision. Increased skull thickness further amplified errors for superficial sources, while its impact on deep sources remained minimal. These findings collectively underscore that high-fidelity representation of skull geometry and tissue properties is essential for the accurate localization of superficial cortical sources, however, it may not sufficiently compensate for the low spatial sensitivity to deep structures, such as the thalamus. This distinction is critical in translational applications, where EEG is used to probe cortical networks (e.g., DMN in psychiatric paradigms) or subcortical circuits (e.g., thalamocortical interactions) and calls for cautious interpretation or methodological supplementation when targeting deep brain regions. The proposed study aimed to reflect realistically expected sources of activity in the rat living brain by using eight dipoles with fixed positions. However, the distance between dipoles and electrodes inherently influence the resulting electric potentials, and the sensitivity to skull inhomogeneity and anisotropy varies accordingly. Consequently, the effects of the skull properties may manifest differently across individual dipoles related to the electrodes, being more pronounced for dipoles located closed to the skull and less evident for deeper sources. To mitigate these factors, we investigated the effects of inhomogeneity and anisotropy on a volumetric grid. We found a strong and specific relationship between the dipole’s most sensitive orientation and the inner surface of the skull, indicating that the closer a dipole is to the skull, the more perpendicular (radial) its most sensitive orientation becomes. The only exception was observed when skull anisotropy was neglected. In this case, the most sensitive dipoles tended to exhibit a parallel (tangential) orientation, particularly in the vicinity of the electrodes. These findings suggest that at least two factors influence the primary direction of the affected dipoles: the distance from anisotropic tissue and the distance from the electrodes. Conclusion In this paper, we investigated the influence of skull inhomogeneity and anisotropy on the forward modeling. The study was carried out on the five cuboid-shaped phantoms differing in the skull material properties and two realistic skull ones with varying anisotropy definitions. Further, we also analyzed the effect of the skull thickness and anisotropy ratios. The findings emphasize the importance of considering skull material properties, including anisotropy and thickness for accurate forward problem solutions. However, some effect that appear critical for the forward solution might have less impact when it comes to the inverse solution. Therefore, future work will be focused on inverse problems to provide more comprehensive understanding of the rat head tissue influence. The emphasis will be put on testing the position-orientation relationship after mitigation of the skull anisotropic properties. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Author contributions D. K., J. L., V. K., C. V., D. K. W., and T. P. established the research objectives, D. K., J. L., V. K., C. V., and D. K. W. contributed to the subsequent development of research questions, D. K. carried out the simulation of forward models in EM software, J. L., F. D., and F. K. contributed to physical phantom development and fabrication, C. V. and V. K. carried out the EEG measurement, J. L. and T. P. provided funding, material resources, and laboratory equipment, D. K. processed the data, created the figures, and took the lead in preparing the manuscript. All authors provided feedback that helped shape the composition of the manuscript. Funding This work was financially supported by the Czech Science Foundation under Grant 23–07578 K. Data availability A dataset containing EEG measurement and rat skull layer thicknesses used to derive the cuboid-shaped phantom Level 2 is available in an online repository on Zenodo at https://doi.org/10.5281/zenodo.17975304. Moreover, the scripts used for the calculation of the error distributions shown in Fig. 7 are publicly available at the following GitHub repository: https://github.com/VlastaKoudelka/skull-leadfield. Additional data or related materials can be shared upon request; interested readers may contact D.K. (xkurat01@vut.cz) or J.L. (lacik@vut.cz) directly. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information David Kuratko, Email: xkurat01@vut.cz. Jaroslav Lacik, Email: lacik@vut.cz. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement A dataset containing EEG measurement and rat skull layer thicknesses used to derive the cuboid-shaped phantom Level 2 is available in an online repository on Zenodo at https://doi.org/10.5281/zenodo.17975304. Moreover, the scripts used for the calculation of the error distributions shown in Fig. 7 are publicly available at the following GitHub repository: https://github.com/VlastaKoudelka/skull-leadfield. Additional data or related materials can be shared upon request; interested readers may contact D.K. (xkurat01@vut.cz) or J.L. (lacik@vut.cz) directly.