Comparison of realistic head modeling methods in EEG source imaging - biomed 2010.

EEG inverse source imaging aims at reconstructing the underlying current distribution in the human brain using potential differences measured non-invasively from the head surface. A critical component of source reconstruction is the head volume conductor model used to reach an accurate solution of the associated forward problem, i.e., the simulation of the EEG for a known current source in the brain. The volume conductor model contains both the geometry and the electrical conduction properties of the head tissues and the accuracy of both parameters has direct impact on the accuracy of the source analysis. This was examined in detail with two different human head models. Two realistic head models derived from an averaged T1-weighted MRI dataset of the Montreal Neurological Institute (MNI) were used for this study. These models were: (1) BEM Model: a four-shell surface-based Boundary Elements (BEM) head model; (2) FDM Model: a volume-based Finite Difference (FDM) model, which allows better modeling accuracy than BEM as it better represents the cortical structures, such as, sulci and gyri in the brain in a three-dimensional head model. How model accuracy description influences the EEG source localizations was studied with the above realistic models of the head. We present here a detailed computer simulation study in which the performances of the two realistic four-shell head models are compared, the realistic MNI-based BEM Model and the FDM Model. As figures of merit for the comparative analysis, the point spread function (PSF) maps and the lead field (LF) correlation coefficients are used. The obtained results demonstrate that a better description of realistic geometry can provide a factor of improvement particularly important when considering sources placed in the temporal or in the occipital cortex. In these situations, using a more refined realistic head model will allow a better spatial discrimination of neural sources.

Threedimensional eeg source reconstruction on high performance computers methodological and computational issues.

This paper presents an original problem solving framework specifically conceived and designed to achieve high performance three-dimensional (3D) reconstruction of the sources of electroencephalographic (EEG) brain activity, named TEBAM (True Electrical Brain Activity Mapping). We describe the integrated framework that has been proposed and developed, specifying TEBAM's design characteristics, implementation and tools interconnections (pipelines). TEBAM relays on patient's specific realistic head modeling for the EEG forward and inverse problem evaluation and is implemented and optimized with a very flexible approach to solve in short time, by means of High Performance Computing resources, the large scale computations needed. Results of 3D True Electrical Brain Activity Mapping can be visualized in TEBAM framework in different multimodal ways, combining the anatomical information with the computed results to give an optimal insight of computation output, relying also on stereographic visualization.

Head modeling for realistic electrical brain activity mapping identification of a multimodal neuroimaging protocol.

Realistic electrical brain activity mapping implies reconstructing and visualizing sources of electrical brain activity within the specific patient's head. This requires the assumption of a precise and realistic volume conductor model of the specific subject's head, i.e., a 3-D representation of the head's electrical properties in terms of shape and electrical conductivities. Source reconstruction accuracy is influenced by errors committed in head modeling. Clinical images, MRI and CT, are used to identify the head structures to be included in the volume conductor head model. Modeling accuracy mainly relies on the correct image-based identification of head structures, characterized by different electrical conductivities, to be included as separate compartments in the model. This paper analyzes the imaging protocols used in clinical practice to define the most suitable procedures for identification of the various head structures necessary to build an accurate head model also in the presence of morphologic brain pathologies. Furthermore, tissues anisotropy is discussed and identified as well. With this work we have identified a protocol for the acquisition of multimodal patient's imaging data for realistic electrical brain activity mapping purposes, able to account for pathological conditions and for head tissues anisotropy.

A FDM anisotropic formulation for EEG simulation.

Accurate head modeling is required to properly simulate bioelectric phenomena in 3-D as well as to estimate the 3-D bioelectric activity starting from superficial bioelectric measurements and 3-D imaging. Aiming to build an accurate and realistic representation of the volume conductor of the head, also the anisotropy of head tissues should be taken into account. In this paper we describe a new finite-difference method (FDM) formulation which accounts for anisotropy of the various head tissues. Our proposal, being based on FDM, derives the head model directly from patient's specific clinical images. We present here the details of the numerical formulation and the method validation by comparing our numerical proposal and known analytical results using a multi-shell anisotropic head model with skull anisotropy. Furthermore, we analyzed also different numerical grid refinement and EEG source characteristics. The comparison with previously developed FDM methods shows a good performance of the proposed method.

Three-dimensional finite-difference EEG forward problem solution on high performance computers.

EEG forward problem solution using numerical head models with the same resolution and geometry as that available from MRI is desirable. This implies dealing with realistic head models of over 2 million elements, for which problem solution has so far been impractical due to issues of computation time and memory. This paper investigates the possibilities given by high performance computing (HPC) to obtain efficient EEG forward problem solution with high resolution head models of over 2 million elements at reduced computation time. In this paper, a finite difference forward problem solution based on HPC is proposed and tested with parallel implementations of different complexity. Solution feasibility with different HPC schemes is analyzed to individuate, by cost-effectiveness, the most appropriate system configurations allowing the best performances. Results indicate that a feasible solution based on a cluster of 8 processors is convenient, obtaining computation times not higher than 2 minutes. Increasing the cluster size above 32 processors gives no significant improvement in the computation times.

Referenced EEG and head volume conductor model: geometry and parametrical setting.

Brain electrical activity effects spread (spatially) over the whole head volume conductor. Electric scalp potentials (EEG) are the measurable evidences of such activity. EEG forward problem solution involves computing the scalp potentials at a finite set of sensor locations for a source configuration in a specified volume conductor model of the head or of part of it (reduced model). The use of reduced models is appealing for computational reasons. The skull could be, according to its conductivity, the natural bound for bioelectrical currents flow. However there are huge uncertainties on actual skull conductivity (i.e., 1/80 or 1/15 of brain conductivity). We show here the limits besides which model reduction is possible preserving EEG simulation accuracy according to the two competitive definitions for skull conductivity. The identified limits involve a proper choice for the EEG reference (Cz) as well as dipole equivalent source characteristics (position and orientation). To this end we adopted realistic test head models extended to different percentages of two reference models (extended to the chin) which differ for skull conductivity set either to 1/80 or 1/15 of brain conductivity.