An evaluation of dipole reconstruction accuracy with spherical and realistic head models in MEG

DeepEZ: A Graph Convolutional Network for Automated Epileptogenic Zone Localization from Resting-State fMRI Connectivity

OBJECTIVE

Epileptogenic zone (EZ) localization is a crucial step during diagnostic work up and therapeutic planning in medication refractory epilepsy. In this paper, we present the first deep learning approach to localize the EZ based on resting-state fMRI (rs-fMRI) data.

METHODS

Our network, called DeepEZ, uses a cascade of graph convolutions that emphasize signal propagation along expected anatomical pathways. We also integrate domain-specific information, such as an asymmetry term on the predicted EZ and a learned subject-specific bias to mitigate environmental confounds.

RESULTS

We validate DeepEZ on rs-fMRI collected from 14 patients with focal epilepsy at the University of Wisconsin Madison. Using cross validation, we demonstrate that DeepEZ achieves consistently high EZ localization performance (Accuracy: 0.88 ± 0.03; AUC: 0.73 ± 0.03) that far outstripped any of the baseline methods. This performance is notable given the variability in EZ locations and scanner type across the cohort.

CONCLUSION

Our results highlight the promise of using DeepEZ as an accurate and noninvasive therapeutic planning tool for medication refractory epilepsy.

SIGNIFICANCE

While prior work in EZ localization focused on identifying localized aberrant signatures, there is growing evidence that epileptic seizures affect inter-regional connectivity in the brain. DeepEZ allows clinicians to harness this information from noninvasive imaging that can easily be integrated into the existing clinical workflow.

Cortical Source Analysis of Event-Related Potentials: A Developmental Approach

Cortical source analysis of electroencephalographic (EEG) signals has become an important tool in the analysis of brain activity. The aim of source analysis is to reconstruct the cortical generators (sources) of the EEG signal recorded on the scalp. The quality of the source reconstruction relies on the accuracy of the forward problem, and consequently the inverse problem. An accurate forward solution is obtained when an appropriate imaging modality (i.e., structural magnetic resonance imaging - MRI) is used to describe the head geometry, precise electrode locations are identified with 3D maps of the sensor positions on the scalp, and realistic conductivity values are determined for each tissue type of the head model. Together these parameters contribute to the definition of realistic head models. Here, we describe the steps necessary to reconstruct the cortical generators of the EEG signal recorded on the scalp. We provide an example of source reconstruction of event-related potentials (ERPs) during a face-processing task performed by a 6-month-old infant. We discuss the adjustments necessary to perform source analysis with measures different from the ERPs. The proposed pipeline can be applied to the investigation of different cognitive tasks in both younger and older participants.

Automated seizure activity tracking and onset zone localization from scalp EEG using deep neural networks

We propose a novel neural network architecture, SZTrack, to detect and track the spatio-temporal propagation of seizure activity in multichannel EEG. SZTrack combines a convolutional neural network encoder operating on individual EEG channels with recurrent neural networks to capture the evolution of seizure activity. Our unique training strategy aggregates individual electrode level predictions for patient-level seizure detection and localization. We evaluate SZTrack on a clinical EEG dataset of 201 seizure recordings from 34 epilepsy patients acquired at the Johns Hopkins Hospital. Our network achieves similar seizure detection performance to state-of-the-art methods and provides valuable localization information that has not previously been demonstrated in the literature. We also show the cross-site generalization capabilities of SZTrack on a dataset of 53 seizure recordings from 14 epilepsy patients acquired at the University of Wisconsin Madison. SZTrack is able to determine the lobe and hemisphere of origin in nearly all of these new patients without retraining the network. To our knowledge, SZTrack is the first end-to-end seizure tracking network using scalp EEG.

Automatic and robust noise suppression in EEG and MEG: The SOUND algorithm

Electroencephalography (EEG) and magnetoencephalography (MEG) often suffer from noise- and artifact-contaminated channels and trials. Conventionally, EEG and MEG data are inspected visually and cleaned accordingly, e.g., by identifying and rejecting the so-called "bad" channels. This approach has several shortcomings: data inspection is laborious, the rejection criteria are subjective, and the process does not fully utilize all the information in the collected data. Here, we present noise-cleaning methods based on modeling the multi-sensor and multi-trial data. These approaches offer objective, automatic, and robust removal of noise and disturbances by taking into account the sensor- or trial-specific signal-to-noise ratios. We introduce a method called the source-estimate-utilizing noise-discarding algorithm (the SOUND algorithm). SOUND employs anatomical information of the head to cross-validate the data between the sensors. As a result, we are able to identify and suppress noise and artifacts in EEG and MEG. Furthermore, we discuss the theoretical background of SOUND and show that it is a special case of the well-known Wiener estimators. We explain how a completely data-driven Wiener estimator (DDWiener) can be used when no anatomical information is available. DDWiener is easily applicable to any linear multivariate problem; as a demonstrative example, we show how DDWiener can be utilized when estimating event-related EEG/MEG responses. We validated the performance of SOUND with simulations and by applying SOUND to multiple EEG and MEG datasets. SOUND considerably improved the data quality, exceeding the performance of the widely used channel-rejection and interpolation scheme. SOUND also helped in localizing the underlying neural activity by preventing noise from contaminating the source estimates. SOUND can be used to detect and reject noise in functional brain data, enabling improved identification of active brain areas.

Combined EEG/MEG can outperform single modality EEG or MEG source reconstruction in presurgical epilepsy diagnosis

We investigated two important means for improving source reconstruction in presurgical epilepsy diagnosis. The first investigation is about the optimal choice of the number of epileptic spikes in averaging to (1) sufficiently reduce the noise bias for an accurate determination of the center of gravity of the epileptic activity and (2) still get an estimation of the extent of the irritative zone. The second study focuses on the differences in single modality EEG (80-electrodes) or MEG (275-gradiometers) and especially on the benefits of combined EEG/MEG (EMEG) source analysis. Both investigations were validated with simultaneous stereo-EEG (sEEG) (167-contacts) and low-density EEG (ldEEG) (21-electrodes). To account for the different sensitivity profiles of EEG and MEG, we constructed a six-compartment finite element head model with anisotropic white matter conductivity, and calibrated the skull conductivity via somatosensory evoked responses. Our results show that, unlike single modality EEG or MEG, combined EMEG uses the complementary information of both modalities and thereby allows accurate source reconstructions also at early instants in time (epileptic spike onset), i.e., time points with low SNR, which are not yet subject to propagation and thus supposed to be closer to the origin of the epileptic activity. EMEG is furthermore able to reveal the propagation pathway at later time points in agreement with sEEG, while EEG or MEG alone reconstructed only parts of it. Subaveraging provides important and accurate information about both the center of gravity and the extent of the epileptogenic tissue that neither single nor grand-averaged spike localizations can supply.

Effects of forward model errors on EEG source localization

Subject-specific four-layer boundary element method (BEM) electrical forward head models for four participants, generated from magnetic resonance (MR) head images using NFT ( www.sccn.ucsd.edu/wiki/NFT ), were used to simulate electroencephalographic (EEG) scalp potentials at 256 recorded electrode positions produced by single current dipoles of a 3-D grid in brain space. Locations of these dipoles were then estimated using gradient descent within five template head models fit to the electrode positions. These were: a spherical model, three-layer and four-layer BEM head models based on the Montreal Neurological Institute (MNI) template head image, and these BEM models warped to the recorded electrode positions. Smallest localization errors (4.1-6.2 mm, medians) were obtained using the electrode-position warped four-layer BEM models, with largest localization errors (~20 mm) for most basal brain locations. When we increased the brain-to-skull conductivity ratio assumed in the template model scalp projections from the simulated value (25:1) to a higher value (80:1) used in earlier studies, the estimated dipole locations moved outwards (12.4 mm, median). We also investigated the effects of errors in co-registering the electrode positions, of reducing electrode counts, and of adding a fifth, isotropic white matter layer to one individual head model. Results show that when individual subject MR head images are not available to construct subject-specific head models, accurate EEG source localization should employ a four- or five-layer BEM template head model incorporating an accurate skull conductivity estimate and warped to 64 or more accurately 3-D measured and co-registered electrode positions.

Diffuse cerebral language representation in tuberous sclerosis complex

INTRODUCTION

Tuberous sclerosis complex (TSC) is a multisystem genetic disorder affecting multiple organs, including the brain, and very often associated with epileptic activity. Language acquisition and development seems to be altered in a significant proportion of patients with TSC. In the present study, we used magnetoencephalography (MEG) to investigate spatiotemporal cerebral language processing in subjects with TSC and epilepsy during a reading semantic decision task, compared to healthy control participants.

METHODS

Fifteen patients with TSC and 31 healthy subjects performed a lexico-semantic decision task during MEG recording. Minimum-norm estimates (MNE) were computed allowing identification of cerebral generators of language evoked fields (EF) in each subject.

RESULTS

Source analysis of the language EF demonstrated early bilateral medial occipital activation (125ms) followed by a fusiform gyrus activation around 135ms. At 270ms post stimuli presentation, a strong cerebral activation was recorded in the left basal temporal language area. Finally, cerebral activations were measured in Wernicke's area followed by Broca's area. The healthy control group showed larger and earlier language activations in Broca and Wernicke's areas compared to TSC patients. Moreover, cerebral activation from Broca's area was greater than activation from Wernicke's area in both groups, but this difference between anterior and posterior regions was smaller in the TSC group. Finally, the activation latency difference between Broca and Wernicke's areas was greater in healthy controls than in TSC patients, which shows that activations in these areas are more serial in control subjects compared to TSC patients in whom activations occur more simultaneously.

CONCLUSIONS

This is the first study to investigate cerebral language pattern in patients with TSC. Compared to healthy controls, atypical neuromagnetic language responses may reflect cerebral reorganization in these patients in response to early epileptogenic activity or presence at birth of multiple brain lesions.

Decreased language laterality in tuberous sclerosis complex: a relationship between language dominance and tuber location as well as history of epilepsy

Nearly 90% of patients with tuberous sclerosis complex (TSC) have epilepsy. Epilepsy surgery can be considered, which often requires a presurgical assessment of language lateralization. This is the first study to investigate language lateralization in TSC patients using magnetoencephalography. Fifteen patients performed a language task during magnetoencephalography recording. Cerebral generators of language-evoked fields (EF) were identified in each patient. Laterality indices (LI) were computed using magnetoencephalography data extracted from the inferior frontal as well as middle and superior temporal gyri from both hemispheres between 250 and 550 ms. Source analysis demonstrated a fusiform gyrus activation, followed by an activation located in the basal temporal language area and middle and superior temporal gyri responses, ending with an inferior frontal activation. Eleven patients (73.3%) had left-hemisphere language dominance, whereas four patients (26.7%) showed a bilateral language pattern, which was associated with a history of epilepsy and presence of tubers in language-related areas.

Language lateralization represented by spatiotemporal mapping of magnetoencephalography

BACKGROUND AND PURPOSE

Determination of hemispheric language dominance is critical for planning epilepsy surgery. We assess the usefulness of spatiotemporal source analysis of magnetoencephalography for determining language laterality.

MATERIALS AND METHODS

Thirty-five patients with epilepsy were studied. The patients performed a semantic word-processing task during MEG recording. Epochs containing language-related neuromagnetic activity were averaged after preprocessing. The averaged data between 250 and 550 ms after stimulus were analyzed by using dynamic statistical parametric mapping. ROIs were obtained in the opercular and triangular parts of the inferior frontal gyrus, superior temporal gyrus, and supramarginal gyrus in both hemispheres. We calculated laterality indices according to 1) dSPM-amplitude method, based on the amplitude of activation in the ROIs, and 2) dSPM-counting method, based on the number of unit dipoles with activation over a threshold in the ROIs. The threshold was determined as half of the maximum value in all ROIs for each patient. A LI ≥0.10 or ≤-0.10 was considered left- or right-hemisphere dominance, respectively; a LI between -0.10 and 0.10 was considered bilateral. All patients underwent an intracarotid amobarbital procedure as part of presurgical evaluation.

RESULTS

The dSPM-counting method demonstrated laterality consistent with the IAP in 32 of 35 patients (91.4%), the remaining 3 (8.6%) demonstrated bilateral language representation, whereas the dSPM-amplitude method showed 18 (51.4%) concordant and 17 (48.6%) bilateral. No laterality opposite to the IAP was found.

CONCLUSIONS

Spatiotemporal mapping of language lateralization with the dSPM-counting method may reduce the necessity for an IAP in as many as 90% of patients.

Evaluation of multiple-sphere head models for MEG source localization

Magnetoencephalography (MEG) source analysis has largely relied on spherical conductor models of the head to simplify forward calculations of the brain's magnetic field. Multiple- (or overlapping, local) sphere models, where an optimal sphere is selected for each sensor, are considered an improvement over single-sphere models and are computationally simpler than realistic models. However, there is limited information available regarding the different methods used to generate these models and their relative accuracy. We describe a variety of single- and multiple-sphere fitting approaches, including a novel method that attempts to minimize the field error. An accurate boundary element method simulation was used to evaluate the relative field measurement error (12% on average) and dipole fit localization bias (3.5 mm) of each model over the entire brain. All spherical models can contribute in the order of 1 cm to the localization bias in regions of the head that depart significantly from a sphere (inferior frontal and temporal). These spherical approximation errors can give rise to larger localization differences when all modeling effects are taken into account and with more complex source configurations or other inverse techniques, as shown with a beamformer example. Results differed noticeably depending on the source location, making it difficult to recommend a fitting method that performs best in general. Given these limitations, it may be advisable to expand the use of realistic head models.

Neuroelectromagnetic forward head modeling toolbox

This paper introduces a Neuroelectromagnetic Forward Head Modeling Toolbox (NFT) running under MATLAB (The Mathworks, Inc.) for generating realistic head models from available data (MRI and/or electrode locations) and for computing numerical solutions for the forward problem of electromagnetic source imaging. The NFT includes tools for segmenting scalp, skull, cerebrospinal fluid (CSF) and brain tissues from T1-weighted magnetic resonance (MR) images. The Boundary Element Method (BEM) is used for the numerical solution of the forward problem. After extracting segmented tissue volumes, surface BEM meshes can be generated. When a subject MR image is not available, a template head model can be warped to measured electrode locations to obtain an individualized head model. Toolbox functions may be called either from a graphic user interface compatible with EEGLAB (http://sccn.ucsd.edu/eeglab), or from the MATLAB command line. Function help messages and a user tutorial are included. The toolbox is freely available under the GNU Public License for noncommercial use and open source development.

Propagation of epileptic spikes reconstructed from spatiotemporal magnetoencephalographic and electroencephalographic source analysis

The purpose of this study is to assess the accuracy of spatiotemporal source analysis of magnetoencephalography (MEG) and scalp electroencephalography (EEG) for representing the propagation of frontotemporal spikes in patients with partial epilepsy. This study focuses on frontotemporal spikes, which are typically characterized by a preceding anterior temporal peak followed by an ipsilateral inferior frontal peak. Ten patients with frontotemporal spikes on MEG/EEG were studied. We analyzed the propagation of temporal to frontal epileptic spikes on both MEG and EEG independently by using a cortically constrained minimum norm estimate (MNE). Spatiotemporal source distribution of each spike was obtained on the cortical surface derived from the patient's MRI. All patients underwent an extraoperative intracranial EEG (IEEG) recording covering temporal and frontal lobes after presurgical evaluation. We extracted source waveforms of MEG and EEG from the source distribution of interictal spikes at the sites corresponding to the location of intracranial electrodes. The time differences of the ipsilateral temporal and frontal peaks as obtained by MEG, EEG and IEEG were statistically compared in each patient. In all patients, MEG and IEEG showed similar time differences between temporal and frontal peaks. The time differences of EEG spikes were significantly smaller than those of IEEG in nine of ten patients. Spatiotemporal analysis of MEG spikes models the time course of frontotemporal spikes as observed on IEEG more adequately than EEG in our patients. Spatiotemporal source analysis may be useful for planning epilepsy surgery, by predicting the pattern of IEEG spikes.

EEG/MEG source imaging: methods, challenges, and open issues

We present the four key areas of research-preprocessing, the volume conductor, the forward problem, and the inverse problem-that affect the performance of EEG and MEG source imaging. In each key area we identify prominent approaches and methodologies that have open issues warranting further investigation within the community, challenges associated with certain techniques, and algorithms necessitating clarification of their implications. More than providing definitive answers we aim to identify important open issues in the quest of source localization.

Mapping the signal-to-noise-ratios of cortical sources in magnetoencephalography and electroencephalography

Although magnetoencephalography (MEG) and electroencephalography (EEG) have been available for decades, their relative merits are still debated. We examined regional differences in signal-to-noise-ratios (SNRs) of cortical sources in MEG and EEG. Data from four subjects were used to simulate focal and extended sources located on the cortical surface reconstructed from high-resolution magnetic resonance images. The SNR maps for MEG and EEG were found to be complementary. The SNR of deep sources was larger in EEG than in MEG, whereas the opposite was typically the case for superficial sources. Overall, the SNR maps were more uniform for EEG than for MEG. When using a noise model based on uniformly distributed random sources on the cortex, the SNR in MEG was found to be underestimated, compared with the maps obtained with noise estimated from actual recorded MEG and EEG data. With extended sources, the total area of cortex in which the SNR was higher in EEG than in MEG was larger than with focal sources. Clinically, SNR maps in a patient explained differential sensitivity of MEG and EEG in detecting epileptic activity. Our results emphasize the benefits of recording MEG and EEG simultaneously.

Dynamic statistical parametric mapping for analyzing ictal magnetoencephalographic spikes in patients with intractable frontal lobe epilepsy

The purpose of this study is to assess the clinical value of spatiotemporal source analysis for analyzing ictal magnetoencephalography (MEG). Ictal MEG and simultaneous scalp EEG was recorded in five patients with medically intractable frontal lobe epilepsy. Dynamic statistical parametric maps (dSPMs) were calculated at the peak of early ictal spikes for the purpose of estimating the spatiotemporal cortical source distribution. DSPM solutions were mapped onto a cortical surface, which was derived from each patient's MRI. Equivalent current dipoles (ECDs) were calculated using a single-dipole model for comparison with dSPMs. In all patients, dSPMs tended to have a localized activation, consistent with the clinically determined ictal onset zone, whereas most ECDs were considered to be inappropriate sources according to their goodness-of-fit values. Analyzing ictal MEG spikes by using dSPMs may provide useful information in presurgical evaluation of epilepsy.

Parallel implementation of the accelerated BEM approach for EMSI of the human brain

Boundary element method (BEM) is one of the numerical methods which is commonly used to solve the forward problem (FP) of electro-magnetic source imaging with realistic head geometries. Application of BEM generates large systems of linear equations with dense matrices. Generation and solution of these matrix equations are time and memory consuming. This study presents a relatively cheap and effective solution for parallel implementation of the BEM to reduce the processing times to clinically acceptable values. This is achieved using a parallel cluster of personal computers on a local area network. We used eight workstations and implemented a parallel version of the accelerated BEM approach that distributes the computation and the BEM matrix efficiently to the processors. The performance of the solver is evaluated in terms of the CPU operations and memory usage for different number of processors. Once the transfer matrix is computed, for a 12,294 node mesh, a single FP solution takes 676 ms on a single processor and 72 ms on eight processors. It was observed that workstation clusters are cost effective tools for solving the complex BEM models in a clinically acceptable time.

Studying the human somatosensory hand area: A new way to compare fMRI and MEG

Valid localization is a prerequisite to study plasticity of the somatosensory cortex in humans. We compared the localizations of left and right thumb and little finger in the primary somatosensory cortex obtained with fMRI and MEG. Representations were investigated in 11 healthy right-handed subjects using echoplanar fMRI and 122-channel MEG together with electric finger stimulation. Activation observed with fMRI was based on an increase in the BOLD signal. Most of the activation clusters (71.1%) were located on the lateral surface of the postcentral gyrus. Representations of thumb and little finger were 17mm apart on average and consistently showed a somatotopic arrangement with the thumb representation inferior, lateral, and anterior to the representation of the little finger. Activation observed with MEG was modelled by equivalent current dipoles. Dipole localization was compatible with an assumed origin of activation within the posterior wall of the central sulcus. The Euclidian distance between corresponding dipoles was 11.5mm on average with deviations from the expected spatial arrangement of 35, 30, and 20% in the x-, y- und z-direction, respectively. Our study demonstrates how relative localization of somatosensory activations can serve as an indicator for localization validity when comparing different methods or studying somatosensory plasticity.

Spatial relationship of source localizations in patients with focal epilepsy: Comparison of MEG and EEG with a three spherical shells and a boundary element volume conductor model

Epilepsy surgery is an option for patients with pharmacoresistant focal epilepsies, but it requires a precise focus localization procedure. Magnetoencephalography (MEG) and electroencephalography (EEG) can be used for analysis of interictal activity. The aim of this prospective study was to compare clusters of source localization results with MEG and EEG using a three spherical shells (3SS) and a boundary element method (BEM) volume conductor model. The study was closed when 100 patients met the inclusion criteria. Simultaneous MEG and EEG were recorded during presurgical evaluation. Epileptiform signals were analyzed using an equivalent current dipole model. Centroids of source localizations from MEG, EEG, 3SS, and BEM in their respective combinations were compared. In a 3SS model, MEG source localizations were 5.6 mm inferior to those obtained by EEG, while in a BEM model MEG source localizations were 6.3 mm anterior and 4.8 mm superior. The mean scattering of source localizations between both volume conductor models was 19.5 mm for EEG and 9.6 mm for MEG. For MEG no systematic difference between BEM and 3SS source localizations was found. For EEG, source localizations with BEM were 5.9 mm posterior and 11.7 mm inferior to those determined using 3SS. No differences were found between the 46 temporal and the 54 extratemporal lobe epilepsy patients. The observed systematic differences of source localizations of epileptic spikes due to the applied source signal modality and volume conductor model should be considered in presurgical evaluation when only one source signal and volume conductor model is available.

A simultaneous EEG–fMRI study of painful electric stimulation

Together with a detailed behavioral analysis, simultaneous measurement of functional magnetic resonance imaging (fMRI) and electroencephalography (EEG) permits a better elucidation of cortical pain processing. We applied painful electrical stimulation to 6 healthy subjects and acquired fMRI simultaneously with an EEG measurement. The subjects rated various stimulus properties and the individual affective state. Stimulus-correlated BOLD effects were found in the primary and secondary somatosensory areas (SI and SII), the operculum, the insula, the supplementary motor area (SMA proper), the cerebellum, and posterior parts of the anterior cingulate gyrus (ACC). Perceived pain intensity was positively correlated with activation in these areas. Higher unpleasantness rating was associated with suppression of activity in areas known to be involved in stimulus categorization and representation (ventral premotor cortex, PCC, parietal operculum, insula) and enhanced activation in areas initiating, propagating, and executing motor reactions (ACC, SMA proper, cerebellum, primary motor cortex). Concordant dipole localizations in SI and ACC were modeled. Using the dipole strength in SI, the network was restricted to SI. The BOLD signal change in ACC was positively correlated to the individual dipole strength of the source in ACC thus revealing a close relationship of BOLD signal and possibly underlying neuronal electrical activity in SI and the ACC. The BOLD signal change decreased in SI over time. Dipole strength of the ACC source decreased over the experiment and increased during the stimulation block suggesting sensitization and habituation effects in these areas.

Computational aspects of the EEG forward problem solution for real head model using finite element method

The real head model has been used for the accurate scalp potential modeling. The realistic shapes of head tissues were derived from a set of 2-D magnetic resonance images (MRI) by extracting surface boundaries for the major tissues such as the scalp, the skull, the cerebrospinal fluid (CSF), the white matter, and the gray matter. From boundary data a 3-D volume generic head model has been constructed and a mesh for an arbitrary complexity head shape has been generated for finite-element method (FEM) modeling. This paper first addresses the use of this realistic finite elements head model to solve the EEG forward problem. The accuracy and computational time of the potential modeling are then examined.

Use of the isolated problem approach for multi-compartment BEM models of electro-magnetic source imaging

The isolated problem approach (IPA) is a method used in the boundary element method (BEM) to overcome numerical inaccuracies caused by the high-conductivity difference in the skull and the brain tissues in the head. Hämäläinen and Sarvas (1989 IEEE Trans. Biomed. Eng. 36 165-71) described how the source terms can be updated to overcome these inaccuracies for a three-layer head model. Meijs et al (1989 IEEE Trans. Biomed. Eng. 36 1038-49) derived the integral equations for the general case where there are an arbitrary number of layers inside the skull. However, the IPA is used in the literature only for three-layer head models. Studies that use complex boundary element head models that investigate the inhomogeneities in the brain or model the cerebrospinal fluid (CSF) do not make use of the IPA. In this study, the generalized formulation of the IPA for multi-layer models is presented in terms of integral equations. The discretized version of these equations are presented in two different forms. In a previous study (Akalin-Acar and Gençer 2004 Phys. Med. Biol. 49 5011-28), we derived formulations to calculate the electroencephalography and magnetoencephalography transfer matrices assuming a single layer in the skull. In this study, the transfer matrix formulations are updated to incorporate the generalized IPA. The effects of the IPA are investigated on the accuracy of spherical and realistic models when the CSF layer and a tumour tissue are included in the model. It is observed that, in the spherical model, for a radial dipole 1 mm close to the brain surface, the relative difference measure (RDM*) drops from 1.88 to 0.03 when IPA is used. For the realistic model, the inclusion of the CSF layer does not change the field pattern significantly. However, the inclusion of an inhomogeneity changes the field pattern by 25% for a dipole oriented towards the inhomogeneity. The effect of the IPA is also investigated when there is an inhomogeneity in the brain. In addition to a considerable change in the scale of the potentials, the field pattern also changes by 15%. The computation times are presented for the multi-layer realistic head model.

An advanced boundary element method (BEM) implementation for the forward problem of electromagnetic source imaging

The forward problem of electromagnetic source imaging has two components: a numerical model to solve the related integral equations and a model of the head geometry. This study is on the boundary element method (BEM) implementation for numerical solutions and realistic head modelling. The use of second-order (quadratic) isoparametric elements and the recursive integration technique increase the accuracy in the solutions. Two new formulations are developed for the calculation of the transfer matrices to obtain the potential and magnetic field patterns using realistic head models. The formulations incorporate the use of the isolated problem approach for increased accuracy in solutions. If a personal computer is used for computations, each transfer matrix is calculated in 2.2 h. After this pre-computation period, solutions for arbitrary source configurations can be obtained in milliseconds for a realistic head model. A hybrid algorithm that uses snakes, morphological operations, region growing and thresholding is used for segmentation. The scalp, skull, grey matter, white matter and eyes are segmented from the multimodal magnetic resonance images and meshes for the corresponding surfaces are created. A mesh generation algorithm is developed for modelling the intersecting tissue compartments, such as eyes. To obtain more accurate results quadratic elements are used in the realistic meshes. The resultant BEM implementation provides more accurate forward problem solutions and more efficient calculations. Thus it can be the firm basis of the future inverse problem solutions.

Reconstruction of 3-D Head Geometry From Digitized Point Sets: An Evaluation Study

In this paper, we evaluate different methods to estimate patient-specific scalp, skull, and brain surfaces from a set of digitized points from the target's scalp surface. The reconstruction problem is treated as a registration problem: An a priori surface model, consisting of the scalp, skull, and brain surfaces, is registered to the digitized surface points. The surface model is generated from segmented magnetic resonance (MR) volume images. We study both affine and free-form deformation (FFD) registration, the use of average models, the averaging of individual registration results, a model selection procedure, and statistical deformation models. The registration algorithms are mainly previously published, and the objective of this paper is to evaluate these methods in this particular application with sparse data. The main interest of this paper is to generate geometric head models for biomedical applications, such as electroencephalography and magnetoencephalographic. However, the methods can also be applied to other anatomical regions and to other application areas. The methods were validated using 15 MR volume images, from which the scalp, skull, and brain were manually segmented. The best results were achieved by averaging the results of the FFD registrations of the database: the mean distance from the manually segmented target surface to a deformed a priori model surface for the studied anatomical objects was 1.68-2.08 mm, depending on the point set used. The results support the use of the evaluated methods for the reconstruction of geometric models in applications with sparse data.

Temporal and spatial congruence of components of motion-onset evoked responses investigated by whole-head magneto-electroencephalography

Motion-onset related components in averaged whole head co-recorded MEG and EEG responses of 24 adults to a low-contrast checkerboard pattern were studied. The aims were to identify these components, to characterize quantitatively their maps and to localize the underlying sources by equivalent-current-dipole (ECD) analyses with a spherical head model.After a weak P1, a large start-elicited negativity arises, comprising the novel N2a (occipital positive and parieto-central negative, peak-latency 141 ms) and the N2 like N2b (bilateral parieto-temporal, 175 ms) component. It is followed by a large positive stop-related component, P2 (156 ms after motion-offset). The corresponding MEG components N2am and N2bm showed bilateral dipole fields with considerable overlap. P1m has a single dipole field around the midline. N2a(m) and N2b(m) can be modelled with two bilateral ECDs with significant different locations. The study shows that accurate mapping and ECD analyses can distinguish two neighbouring areas of the visual cortex, 21+/-4 (SE) mm separated, which activities are reflected in both spatio-temporally closely related N2(m) components. N2a(m) and N2b(m) originate in the extrastriate cortex, possibly close to or in V3/V3A and MT/V5 respectively. Motion-evoked activity in (near) V3/V3A is novel on the basis of EEG data.

Localization of auditory N1 in children using MEG: source modeling issues

Techniques for localizing auditory (AEF) sources are a topic of on-going discussion and this is particularly pertinent in pediatric research. Smaller head sizes are: (1) subject to bilateral temporal lobe source interference from both temporal lobes; and (2) further from MEG sensors resulting in poorer signal-to-noise ratios. An additional consideration in children is that the components of the AEF have distinct contributions along the development spectrum resulting in an ever-changing morphology for the pediatric AEF. These factors present a complicated picture for dipole fitting and raise the question of the most effective fitting strategy. We examined the AEF localizations in five children from 151, 70 and 47 MEG channels of data. We found evidence that bilateral source interaction could result in localization errors along the medial-lateral axis of up to 1 cm. We suggest that any modeling strategy needs to sufficiently account for this interaction and more precise models allowing for multiple sources need to be developed.

The 3D topography of MEG source localization accuracy: effects of conductor model and noise

OBJECTIVE

To evaluate the effect that different head conductor models have on the source estimation accuracy of magnetoencephalography (MEG) under realistic conditions.

METHODS

Magnetic fields evoked by current dipoles were simulated using a highly refined 3-layer realistically shaped conductor model. Noise from a real MEG measurement was added to the simulated fields. Source parameters (location, strength, orientation) were estimated from the noisy signals using 3 spherically symmetric models and several one- and 3-layer realistically shaped boundary-element models. The effect of different measurement sensors (gradiometers, magnetometers) was also tested.

RESULTS

The noise typically present in brain signals masked the errors due to the different conductor models so that in most situations the models gave comparable results. Active cortical areas around the vertex and in the temporal, frontoparietal, and occipital regions were typically found with 2-4 mm accuracy, whereas source localization in several anterior frontal lobe and deep brain structures yielded errors exceeding 2 cm. Localization in anterior frontal regions may benefit most from the use of realistically shaped models.

CONCLUSIONS

The traditionally used sphere model is an adequate model for most research purposes. Any means that increase the signal-to-noise ratio are of highest importance in attempting to improve the source estimation accuracy.

Electrical impedance tomography of human brain function using reconstruction algorithms based on the finite element method

Electrical impedance tomography (EIT) is a recently developed technique which enables the internal conductivity of an object to be imaged using rings of external electrodes. In a recent study, EIT during cortical evoked responses showed encouraging changes in the raw impedance measurements, but reconstructed images were noisy. A simplified reconstruction algorithm was used which modelled the head as a homogeneous sphere. In the current study, the development and validation of an improved reconstruction algorithm are described in which realistic geometry and conductivity distributions have been incorporated using the finite element method. Data from computer simulations and spherical or head-shaped saline-filled tank phantoms, in which the skull was represented by a concentric shell of plaster of Paris or a real human skull, have been reconstructed into images. There were significant improvements in image quality as a result of the incorporation of accurate geometry and extracerebral layers in the reconstruction algorithm. Image quality, assessed by blinded subjective expert observers, also improved significantly when data from the previous evoked response study were reanalysed with the new algorithm. In preliminary images collected during epileptic seizures, the new algorithm generated EIT conductivity changes which were consistent with the electrographic ictal activity. Incorporation of realistic geometry and conductivity into the reconstruction algorithm significantly improves the quality of EIT images and lends encouragement to the belief that EIT may provide a low-cost, portable functional neuroimaging system in the foreseeable future.

Realistic human head model for EEG from both the geometry and conductivity aspects

This paper describes modelling and simulation of a had model which incorporates both the geometries and conductivities of the human head. It focuses on the inclusion of tissue conductivity inhomogeneity in a realistically-shaped head model, and investigates the impact of this inclusion on the potential distribution within the model. The result show that the impact, which has been neglected in realistic head models so far, is significant.

A multivariate, spatiotemporal analysis of electromagnetic time-frequency data of recognition memory

Electromagnetic indices of "fast" (above 12 Hz) oscillating brain activity are much more likely to be considerably attenuated by time-averaging across multiple trials than "slow" (below 12 Hz) oscillating brain activity. To the extent that both types of oscillations represent the activity of temporally and topographically separable neural populations, time averaging can cause a loss of brain activity information that is important both conceptually and for multimodal integration with hemodynamic techniques. To address this issue for recognition memory, simultaneous electroencephalography (EEG) and whole-head magnetoencephalography (MEG) recordings of explicit word recognition from 11 healthy subjects were analyzed in two different ways. First, the time course of neural oscillations ranging from theta (4.5 Hz) to gamma (42 Hz) frequencies were identified using single-trial continuous wavelet transforms. Second, traditional analyses of amplitude variations of time-averaged EEG and MEG signals, event-related potentials (ERPs), and fields (ERFs) were performed and submitted to distributed source analyses. To identify data patterns that covaried with the difference between correctly recognized studied (old) words and correctly rejected nonstudied (new) words, a multivariate statistical tool, partial least squares (PLS), was applied to both types of analyses. The results show that ERPs and ERFs are mainly displaying those neural indices of recognition memory that oscillate in the theta (4.5-7.5 Hz), alpha (8-11.5), and to some extent in the beta1 (12-19.5 Hz) frequency range. The sources of the ERPs/ERFs were in good agreement with the topography of theta/alpha/beta 1 oscillations in being confined to the anterior temporal lobe at 400 ms and being distributed across temporal, parietal, and occipital areas between 500 and 700 ms. Gamma oscillations covaried either positively or negatively with theta/alpha/beta1 oscillations. A positive covariance, for instance, was detected over left anterior temporal sensors as early as 200-350 ms and is compatible with studies in rodents showing that gamma and theta oscillations emerge together out of the interaction of the hippocampus and the entorhinal and perirhinal cortices. Fast beta oscillations (20-29.5 Hz), on the other hand, did not strongly covary with slow oscillations and were likely to arise from neural populations not adequately represented in ERPs/ERFs. In summary, by providing a more comprehensive description of electromagnetic signals, time-frequency data are of potential benefit for integrating electrophysiological and hemodynamic indices of brain activity and also for integrating human and animal electrophysiology.

Multidipole analysis of simulated epileptic spikes with real background activity

This simulated magnetoencephalographic study was designed to determine the variability in source parameters with real subject background activity when applying multidipole spatial-temporal dipole analyses, for which the correct model was compared with undermodeled and overmodeled cases. The simulated sources were created from patches of the cortical surface of each subject's MRI. One- and two-source frontal lobe spikes were generated in two cortical regions seen commonly in frontal lobe epilepsy patients tested at our site (orbital frontal and premotor cortex). In general, the modeling results were adequate for the correct model order and the correct model order plus one. In addition, if the localization error was less than 10 mm from the simulated source, the peak latency of the spike and orientation were very reliable, but the peak amplitude was not. The additional source in the overmodeled condition, on the other hand, was not localized reliably across the different epochs within subjects. The results suggest that consistency of the spike localization and inconsistency of other sources will allow one to determine reliably the appropriate model order in real data, and therefore determine single and multifocal spike generators.

Time-coherent expansion of MEG/EEG cortical sources

In this study we estimated the spatial extent of cortical areas of time-coherent activity using the inverse problem in magneto/electroencephalography (MEEG). The model discussed here uses classical regularization tools in order to force the inverse solution to be piecewise coherent. First, the cortex was seeded by focal dipolar sources. Then, a time-coherent expansion (TCE) onto the cortical surface was performed in order to obtain surface source models composed of patches with uniform current density. Patches represent extended cortical regions with one single time course per active area. Results obtained from synthetic data show that using the TCE method is relevant even with a low signal-to-noise ratio, although the final estimation is often slightly biased. We applied the TCE method to evoked magnetic fields obtained after electrical stimulation of fingers in order to estimate the somatotopic cortical maps of the primary somatosensory cortex.

Detecting functional asymmetries through the dipole moment of magnetoencephalography

To assess the accuracy of magnetoencephalography (MEG) as a tool for quantitative detection of neuronal activity, the dipole moment was estimated at N20m of somatosensory evoked fields (SEFs) produced by median nerve stimulation. Neurologically stable patients were examined twice within 2 weeks. Since the estimated moment values of the two examinations should be essentially the same, we assessed the margin of error for our measurement system. The results showed that a change of more than 5.2 nAm is statistically significant (p=0.05; n=91). The patients were classified as without or with functional asymmetries by measuring the conventional cerebral blood flow (CBF) with single photon emission CT (SPECT), and the dipole moment difference between hemispheres was measured. Hemispheric moment differences were statistically larger for the group with CBF laterality, indicating consistency between conventional CBF results and the moment measurements as a group. Moreover, MEG was able to detect more functional asymmetries than CBF study. The dipole moment provided a reliable quantitative index of cortical response to somatosensory stimulus, and the moment measurement thus holds promise as a clinical tool for direct quantification of cortical response.