Improving source reconstructions by combining bioelectric and biomagnetic data
Introduction
Source reconstructions combining bioelectric and biomagnetic data promise to benefit from the advantages of both modalities (Cohen and Cuffin, 1979; Cohen and Cuffin, 1983; Cohen and Cuffin, 1987; Cohen and Cuffin, 1991; Cohen et al., 1990; Lopes da Silva et al., 1991; Mauguiere, 1992). Electroencephalographic (EEG) measurements can be carried out with an optimized electrode arrangement and provide nearly equal sensitivities for tangentially- and radially-oriented sources. Due to this unspecific sensitivity distribution EEG data often suffer from a limited signal-to-noise-ratio (SNR) and exhibit rather complex field structures. Magnetoencephalographic (MEG) gradiometer systems have an increased sensitivity for tangential superficial sources, leading to an improved SNR and a larger specificity for this class of generators. On the other hand this means that MEG sensor arrays are more or less blind to (quasi-)radial neural current components and deep sources, leading to a reduced complexity of the measured field patterns. Therefore, a combination of both complementary methods should be able to reveal radial dipole components and stabilize the reconstruction of tangential sources by an increased information content and an improved overall SNR.
In order to combine both modalities in a unified framework, different problems have to be solved:
- •
The different measures have to be transformed to a common basis. This is done by referencing each sensor to its individual noise statistics (Greenblatt, 1995; Pflieger et al., 1998). So every measurement channel contributes with its statistical relevance to the ensuing evaluation procedures. One method to automatically determine the noise level of each sensor is to use the standard deviation of a fraction of the smallest signal levels of the total latency range, e.g. the smallest 20%. This method of course requires about 20% signal-free (e.g. pre-trigger) samples in the measurement.
- •
For unified reconstruction algorithms, a common volume conductor model has to be used. EEG data strongly depend on the head's shape and its conductivities (Cuffin, 1990). For example, in volume conductor models with one compartment only, the measured signals are inversely proportional to the conductivity. MEG signals in the spherical volume conductor approximation are not at all affected by the conductivity. With more realistically-shaped volume conductor models (e.g. boundary element method (BEM) models) the MEG data show only a weak dependence on the electric properties of the compartments. Furthermore the individual real (in-vivo) tissue conductivities are not well known (Geddes and Baker, 1963). We have thus chosen latencies where single dominantly tangential dipoles can explain the measured data very well for both modalities and used them for matching the conductivities of the volume conductor model. Thus the magnetic data are used for calibrating the electric conductivities via a common scaling factor that keeps the relative conductivities of the model compartments (brain, skull, skin) constant.
- •
MEG reconstructions with realistic volume conductor models tend to overemphasize quasi-radial current components due to their very small gain (Menninghaus and Lütkenhöner, 1995). Therefore, dipole regularization techniques have to be introduced in order to limit or suppress these low-gain components at least in single-modality MEG evaluations. In EEG or combined EEG/MEG examinations, the electric data are expected to reduce these effects due to their nearly isotropic orientational sensitivity distributions (Fuchs et al., 1998b).
For testing the methods described above in view of their spatial resolution, simulations with different electrode and magnetometer/gradiometer set-ups were used with a 3 spherical shells volume conductor model. White noise, uncorrelated across sensors, was added to calculated field distributions to get statistically relevant results for different dipole depths and signal-to-noise-ratios. Equivalent dipolar sources were then fitted with single and combined modalities.
Evoked somatosensory measurements from electric medianus nerve stimulation with simultaneously recorded 31 electrodes EEG (SEP) and 31 channels MEG (SEF) (Buchner et al., 1994) were used to test combined evaluations with real data. The volume conductor was modeled by a BEM consisting of 3 compartments. The head/brain compartments for the BEM model were semi-automatically segmented, and triangulated from magnetic resonance (MR) images (Wagner et al., 1995). Single equivalent dipoles, cortically constrained deviation scans (Fuchs et al., 1994; Fuchs et al., 1998a), and spatiotemporal dipole models (Scherg and von Cramon, 1985; Mosher et al., 1992) were used to compare single modality and merged evaluations.
Section snippets
Signal-to-noise-ratio transformation
In order to combine the different measures of electric and magnetic data, both have to be converted to a common basis. Using their signal-to-noise-ratios (SNRs) (Greenblatt, 1995; Pflieger et al., 1998) all sensor or sensor group signals are processed in the ensuing reconstruction algorithms according to their statistical significances. A channel-wise SNR transformation can be utilized by determination of the noise amplitude ni of each channel, i, from signal-free (e.g. pre-trigger) latency
Simulations
First the mean SNRs were plotted against the depths of the dipoles. Thus the depth sensitivities of the different single and combined modality set-ups can be studied by analyzing the reduced SNRs, which are equivalent to the relative system sensitivities. The mean SNR values of the most superficial dipole position were used for normalization. After performing single dipole fits, the mean mislocalizations (averaged spatial distances between true and fitted dipole positions) were plotted as a
Conclusions
A framework for combining bioelectric and biomagnetic data on the basis of a signal-to-noise ratio transformation of the data was presented. Common realistic volume conductor models were used and a new regularization approach for the source reconstruction algorithms was introduced, using the partial variances that can be explained by the principal dipole components. In order to test and verify the methods, extensive simulations were carried out. Normalized sensitivity and dipole resolution
Unlinked reference
Oostendorp and van Oosterom, 1989
References (41)
- et al.
Postcentral origin of P22: evidence from source reconstruction in a realistically shaped head model and from a patient with a postcentral lesion
Electroenceph. clin. Neurophysiol.
(1996) - et al.
Demonstration of useful differences between magnetoencephalogram and electroencephalogram
Electroenceph. clin. Neurophysiol.
(1983) - et al.
Two bilateral sources of the late AEP as identified by a spatio-temporal dipole model
Electroenceph. clin. Neurophysiol.
(1985) - Ben-Isreal, A. and Greville, T.N.E. Some topics in generalized inverses of matrices. In: M.Z. Nashed (Ed.), Generalized...
- et al.
Source analysis of median nerve and finger stimulated somato-sensory evoked potentials: multichannel simultaneous recording of electric and magnetic fields combined with 3D-MR tomography
Brain Topogr.
(1994) - et al.
Origin of P16 median nerve SEP component identified by dipole source analysis – subthalamic or within the thalamo-cortical radiation?
Exp. Brain. Res.
(1995) - et al.
Comparison of the magnetoencephalogram and electroencephalogram
Electroenceph. clin. Neurophysiol.
(1979) - et al.
A method for combining MEG and EEG to determine the sources
Phys. Med. Biol.
(1987) - et al.
MEG versus EEG localization test using implanted sources in the human brain
Ann. Neurol.
(1990) - et al.
EEG versus MEG localization accuracy: theory and experiment
Brain Topogr.
(1991)
Effects of head shape on EEGs and MEGs
IEEE Trans. Biomed. Eng.
A complete linear discretization for calculating the magnetic field using the boundary element method
IEEE Trans. Biomed. Eng.
Improved method for computation of potentials in a realistic head shape model
IEEE Trans. Biomed. Eng.
Coordinate system matching for neuromagnetic and morphological reconstruction overlay
IEEE Biomed. Eng.
The specific resistance of biological material, a compendium of data for the biomedical engineer and physiologist
Med. Biol. Eng.
On bioelectric potentials in an inhomogeneous volume conductor
Biophys. J.
Cited by (209)
Standardized hierarchical adaptive Lp regression for noise robust focal epilepsy source reconstructions
2024, Clinical NeurophysiologyThe effects of peeling on finite element method -based EEG source reconstruction
2024, Biomedical Signal Processing and ControlElectromagnetic source imaging predicts surgical outcome in children with focal cortical dysplasia
2023, Clinical NeurophysiologySEEG-RF for revealing and treating Geschwind syndrome's epileptic network: A case study
2023, Epilepsy and Behavior ReportsCan individually targeted and optimized multi-channel tDCS outperform standard bipolar tDCS in stimulating the primary somatosensory cortex?
2023, Brain StimulationCitation Excerpt :Leadfield computation: Leadfields were computed for both EEG and MEG using SimBio-NeuroFEM2 for the final step in forward modeling. EEG and MEG leadfield bases [49] and an algebraic multigrid preconditioned conjugate gradient (AMG-CG) [55] solver, proven to be stable with regard to tissue conductivity inhomogeneity and anisotropy [54,55], were used to achieve high computational speed [53]. Skull conductivity calibration: The combined EEG/MEG reconstruction of the P20/N20 S1 component is interlinked with the calibration of skull conductivity.
Source localization of ictal SEEG to predict postoperative seizure outcome
2022, Clinical NeurophysiologyCitation Excerpt :For both bandpass and FTT methods, noise was estimated from the standard deviation of the 50 percent smallest signal values. Rotating dipoles were computed from bandpass-filtered SEEG (bandpass dipole) and the FFT-derived frequency bin of interest (FFT dipole) using previously described methods (Fuchs et al., 1998, Fuchs et al., 1996). Current density was calculated from the FFT-derived frequency bin of interest using standardized low resolution tomographic analysis (sLORETA) (Pascual-Marqui, 2002) and sLORETA-weighted accurate minimum norm (SWARM) (Wagner et al., 2007).