Development of realistic head models for electromagnetic source imaging of the human brain

Abstract

Abstract-In this work, a methodology is developed to solve the forward problem of electromagnetic source imaging using realistic head models. For this purpose, first segmentation of the 3 dimensional MR head images is performed. Then triangular, quadratic meshes are formed for the interfaces of the tissues. Thus, realistic meshes, representing scalp, skull, CSF, brain and eye tissues, are formed. At least 2000 nodes for the scalp and 5000 for the cortex are needed to obtain reasonable geometrical approximation. Solution of the forward problem using our previous Boundary Element Method (BEM) formulation with quadratic elements remains to be made.Keywords -BEM, realistic head model, segmentation, mesh generation. I. INTRODUCTIONElectrical activities of the human brain due to body functions can be measured with electrodes placed on the scalp (EEG) and with magnetic sensors (MEG) placed near the scalp surface. The representation of electrical activity of the brain using electrical and magnetic measurements is called electromagnetic source imaging (EMSI). The source of an electrical activity is usually modeled by electrical dipoles and the purpose of EMSI is to obtain information about the spatiotemporal behavior of these dipoles. An essential part of obtaining EMSIs is the solution of the electric and magnetic fields (the forward problem) for a given dipoleassuming a head model. The solution of the inverse problem (i.e., given the measured data, finding the location and direction of dipoles) is based on the comparison of the measured and calculated fields. To increase accuracy in EMSIs, the human head must be modeled accurately. The purpose of this study is twofold: 1) to obtain an accurate head model, 2) to solve the forward problem of EMSI for this model.In the earliest studies, head models with simple geometries having analytical solutions for a dipole inside the conductor model were used. The simplest head model is the homogeneous sphere. Other homogenous head models that may represent the head shape are prolate spheroid (eggshape). In order to represent layers like skull, scalp and cerebrospinal fluid (CSF), concentric and eccentric spheres models were used. Such models also have analytical solutions for a dipole inside conductor model [1]. . Thereafter a coarsening algorithm is used to represent the tissues with less number of triangles [7]. Finally, the resulting mesh is corrected topologically [8]. Up to this step linear elements are used. After obtaining the coarse mesh, the elements are converted to quadratic elements using the original segmentation data.In this study, the segmentation and mesh generation algorithms are explained. Meshes created for cortex, white matter, scalp and skull meshes are presented. The BEM formulation [9] which employs triangular, quadratic, isoparametric elements will be used to solve the forward problem of EMSI. II. SEGMENTATIONSegmentation is a process of classifying elements having the same properties in one group. In this work, segmentation of scalp, skull, CSF, eyes, gray m...

Segmentation of skull and scalp in 3‐D human MRI using mathematical morphology

“…The forward problem is then solved using a finite element method. Akahn et al [2001] also use PD and T 1 images to segment skull, scalp, CSF, eyes, and white and gray matter with a hybrid algorithm that uses snakes, region growing, morphological operations, and thresholding. While these approaches are attractive, and the use of PD images will lead to improved segmentation of skull, in the vast majority of brain imaging studies only high-resolution T 1 -weighted images are collected.…”

Investigation of optimal parameters for finite element solution of the forward problem in magnetic field tomography based on magnetoencephalography

“…However, these aspects are rarely covered in the current literature, especially with respect to solution errors and accuracy of FE parameters used for simulations. Human brain is an extremely complex structure for FE modelling both in terms of geometric and material parameters [4,5]. Although accurate incorporation of these has obvious effects on FE solution accuracy, it needs to be balanced against computational time and complexity involved.…”

Multi‐atlas and label fusion approach for patient‐specific MRI based skull estimation

“…In Akahn et al (8), the authors use T1-weighted and proton density images to segment the skull and other tissues with a hybrid algorithm that uses snakes, region growing, morphological operations, and thresholding. The method described in Dogdas et al (9) finds the inner and outer skull boundaries in T1-weighted images using thresholding and morphological operations, and mask the results with the scalp and brain segmentations to ensure closed and nonintersecting skull boundaries.…”