Dear Colleagues,

It is a pleasure to present the special issue of the International Journal of Bioelectro-magnetism devoted to the multimodal integration of EEG, MEG and fMRI. In this issue we would like to present the state-of-the-art of some major research groups in the world working on the theme of the multimodal integration of EEG, MEG and fMRI. Such a theme is important in the neuroscience field nowadays, since there is a large availability of different kinds of electrical, magnetic and hemodynamic “brain scanners”. These scanners return images of the “working brain” under different perspectives and each one present a particular spatio-temporal resolution. Hence, it is of interest to survey how part of the major research groups in this context use the multimodal data to improve the quality of the estimated source activity. The papers presented here can be classified under three principal classes. The first one uses the fMRI data in the context of improving the  localization of a set of discrete sources, the second one uses fMRI data in the context of the estimation of distributed neural sources and the third one addresses theoretically the issue of the multimodal integration.

In the first category, the work by Baillet, Richard M. Leahy, Singh, Shattuck and Mosher addresses the theme of assessing the activity of particular structures of the brain (namely the supplementary motor area) during the motor preparation and the execution with high-order source models derived by MEG data and the use of fMRIs. Authors concluded stressing the aid that fMRI can furnish in terms of localization and spatial extension of the computed sources. Another work in this issue deals with the use of fMRI information for localization of cerebral sources of activity with MEG ( Korvenoja, Aronen and Ilmoniemi). In this study, comparisons of the differences of fMRI and MEG for different experimental paradigms involving somatosensory evoked responses and visual motion processing are made. Authors concluded that even when the experimental set ups in fMRI and MEG were slightly different, a similar activation pattern could be seen with both methods.

The paper by Schwartz, Liu, Bonmassar and Belliveau compared two inverse approaches useful for the estimation of neural sources with discrete and distributed source models. In this simulation study, they compared the generated simulated data using realistic source distributions constrained to the cortical surface with time courses of activation. They examined single and multiple sources with various amounts of temporal overlap and temporal correlation. They suggest that linear estimation provides more accurate temporal information especially when several sources are simultaneously activated with respect to the spatio-temporal discrete localization model. However, they observed more focal solutions with the discrete model than with the distributed one. Finally, they proposed a combined approach to exploit the greater spatial accuracy of the discrete spatio-temporal model within the framework of the linear estimation approach. A methodological approach considering both discrete and distributed solutions with the inclusion of fMRI constraints were pursued also by Wagner and Fuchs. In their work they pointed out how dipole fits can benefit from fMRI constraints, by using the hotspots as seed point for the source localization. However, the presence of spatially unconstrained dipoles are necessary to account for remaining activity not detected by the fMRI scanner. They found also that current density reconstructions react upon fMRI constraints in two ways: activity in the vicinity of fMRI hotspots is bundled, while the remaining activity can be localized correctly if its field distribution cannot be generated from sources within the hotspots, and if the fMRI constraint is imposed softly. The use of fMRI constraints in the estimation of distributed source currents is proposed in the paper by Babiloni F, Babiloni C, Carducci, Angelone, Del Gratta, Romani, Rossini, and Cincotti . Two methods for the modeling of human cortical activity by using combined high-resolution electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) data are presented. Hemodynamic responses of the investigated cortical areas as derived from block-design and event-related functional Magnetic Resonance Imaging (fMRI) are used as a priors in the resolution of the linear inverse problem for the estimation of the cortical activity. The contribution of Gonzalez Andino, Blanke, Lantz, Thut, and Grave de Peralta starts discussing some alternatives to integrate functional information as constraints for the inverse solution. Concrete examples of situations where functional images substantially diverge from electrophysiological methods are presented by these Authors to promote the discussion about the most reasonable alternatives to combine these image modalities. The main conclusion of this paper is that integration of functional modalities into the solution of the NIP should be cautiously considered until a more tight coupling between BOLD effects and electrophysiological measurements could be established.

The paper by Pflieger and Greenblatt has a theoretical nature and it addresses the issues regarding the multimodal integration of MEG and EEG with fMRI signals. In particular, they propose a  linear spatial estimator that maximizes the empirical coupling of the estimated MEG or EEG source activity as driven by local BOLD signal, and a nonlinear dynamic transform that maximizes the coupling of BOLD signal as driven by the estimated MEG or EEG signal. They suggest also that the latter transformation can be the basis for fMRI statistical parametric maps that couple more tightly with neuronal activity compared with task-derived maps. The paper by Trujillo-Barreto, NJ, Martínez-Montes, E, Melie-García, L and Valdés-Sosa, addresses a new method for EEG/MEG and fMRI data fusion, mainly based on a linear model for each kind of measurements, assuming that the variability of the estimated activation in all cases (variance and covariance matrix) is essentially the same, except for a scaling factor. The Point Spread Function (PSF) for the new model is computed, and the results are compared with methods that use only electric measurements. The paper shows that the new methodology has a superior performance according to most of the quality measures used to characterize electrophysiological tomographic techniques. The use of such a method for multimodal EEG/MEG and fMRI fusion is illustrated in the analysis of a somatosensory MEG-fMRI experiment.

We hope that this special issue can be of some utility to the scientific community at least to stimulate the debate about multimodal fMRI-EEG-MEG integration. In this respect, we think that the tribune of International Journal of BioElectromagnetism is suitable to host a fruitful discussion and relevant comments on this issue. Furthermore, we would like to thank sincerely all the Authors that kindly contribute to this editorial trial,  for the excellent quality of their papers.

In conclusion, the scenario that comes out from this overview seems to indicate that the multimodal integration of EEG, MEG and fMRI will play a key role in the future of human brain mapping. Several important research teams worldwide are currently working on the issue of multimodal integration with different methodological approaches and some theoretical breakthroughs can be expected in the near future to further clarify the physiologic link between brain hemodynamic and electrical/magnetic responses. Indeed, the methodological approaches presented here seems ready to be included in the arsenal of computational methods used by neuroscientists to further elucidate the “working brain”.

Fabio Babiloni, Claudio Babiloni, Filippo Carducci and Febo Cincotti

Rome 3 April 2001

 
Jaakko Malmivuo

Tampere 3 April 2001