ORIGINAL PAPER Connecting Mean Field Models of Neural Activity to EEG and fMRI Data Ingo Bojak • Thom F. Oostendorp • Andrew T. Reid • Rolf Ko ¨tter Received: 3 August 2009 / Accepted: 11 March 2010 / Published online: 4 April 2010 Ó Springer Science+Business Media, LLC 2010 Abstract Progress in functional neuroimaging of the brain increasingly relies on the integration of data from complementary imaging modalities in order to improve spatiotemporal resolution and interpretability. However, the usefulness of merely statistical combinations is limited, since neural signal sources differ between modalities and are related non-trivially. We demonstrate here that a mean field model of brain activity can simultaneously predict EEG and fMRI BOLD with proper signal generation and expression. Simulations are shown using a realistic head model based on structural MRI, which includes both dense short-range background connectivity and long-range spe- cific connectivity between brain regions. The distribution of modeled neural masses is comparable to the spatial resolution of fMRI BOLD, and the temporal resolution of the modeled dynamics, importantly including activity conduction, matches the fastest known EEG phenomena. The creation of a cortical mean field model with anatom- ically sound geometry, extensive connectivity, and proper signal expression is an important first step towards the model-based integration of multimodal neuroimages. Keywords Mean field model  Volume conductor model  Multimodal imaging  EEG  fMRI BOLD Introduction Many different non-invasive functional neuroimaging modalities are available. However, linking their data to underlying brain states is far from trivial. All non-invasive modalities are sensitive only to the coherent activity of large groups of neurons. Furthermore, hemodynamic methods with high spatial resolution, like PET and fMRI BOLD, are unable to follow activity on a millisecond temporal scale, whereas methods with high temporal resolution, like EEG and MEG, suffer from the ill-posedness of the associated inverse problem. There are three principle approaches to multi-modal integration (Horwitz and Poeppel 2002). First, converging evidence: a result is compared to a previous one from another modality, either via the literature or by per- forming consecutive measurements, e.g., (Disbrow et al. 2001). However, such comparisons are limited by our lack of knowledge about the relations between the neural gen- erators of different modalities. Second, direct data fusion:a unified statistical correlation analysis is performed, which however still assumes that the different data sets have the same neural generators (Dale and Halgren 2001). This approach is also problematic, since regions of activity in different modalities are often found to be non-overlapping, and their statistical correlations indicate regions which are not prominent individually. See for example Figs. 4 and 6 in (Schulz et al. 2004). Third, computational neural mod- eling, which is the approach we present here: all the dif- ferent data modalities are related to one underlying model of the neural generators. One can then expect that discrep- ancies between data and model will lead to systematic improvements of the model. Further, successful fits of data can be meaningfully interpreted through the model, e.g., specific variations of a model parameter may point to some underlying process known to affect this parameter. This is one of several papers published together in Brain Topography on the ‘‘Special Topic: Cortical Network Analysis with EEG/MEG’’. I. Bojak  T. F. Oostendorp (&)  A. T. Reid  R. Ko ¨tter Donders Institute for Brain, Cognition and Behaviour, Centre for Neuroscience, Radboud University Nijmegen (Medical Centre), P.O. Box 9101//126, 6500 HB Nijmegen, The Netherlands e-mail: T.Oostendorp@donders.ru.nl 123 Brain Topogr (2010) 23:139–149 DOI 10.1007/s10548-010-0140-3