Improvement of Source Localization by Dynamical Systems Based Modeling (DSBM) Christian Uhl*, Axel Hutt*, and Frithjof Kruggel* Summary: Recently, we have proposed a new concept for analyzing EEG/MEG data (Uhl et al. 1998), which leads to a dynamical systems based mod- eling (DSBM) of neurophysiological data. We report the application of this approach to four different classes of simulated noisy data sets, to investi- gate the impact of DSBM-filtering on source localization. An improvement is demonstrated of up to above 50% of the distance between simulated and estimated dipole positions compared to principal component filtered and unfiltered data. On a noise level on which two underlying dipoles cannot be resolved from the unfiltered data, DSBM allows for an extraction of the two sources. Key words: EEG/MEG; Dipole source modeling; Signal filter; Signal reconstruction; Dynamical systems. Introduction A major goal of the analysis of electro- and magnetencephalographic (EEG/MEG) data is the local- ization of brain activity during cognitive tasks. Source modeling in terms of fitting position, orientation and strength of dipole sources (Brenner et al. 1975; Scherg and von Cramon 1986; Mosher et al. 1992) has become a valuable tool for understanding brain functions. Despite the success of dipole modeling, there are still open questions concerning dipole source localiza- tion, such as the definition of components (Rösler 1982) and corresponding time intervals for a basis of the fitting algorithm, and such as the choice of constraints to reduce the ambiguities of the inverse electromagnetic problem. Although the high temporal resolution of EEG/MEG measurements is one of the major advantages of this mo- dality, conventional dipole modeling is not based on temporal aspects for an estimation of spatial parameters. In a recent paper (Uhl et al. 1998) we proposed a new con- cept for analyzing EEG/MEG data aiming at a recon- struction of the signal and its dynamics. The approach is based on a simultaneous fit of spatial and temporal pa- rameters to given data sets, which we will call in follow- ing "dynamical systems based modeling" (DSBM). The reconstruction is achieved by minimizing a cost function considering signal representation - as it is done in princi- pal component analysis (PCA) (Donchin and Heffley 1978) - and dynamic interactions. In our view, this may lead to more objective criteria for the characterization of EEG/MEG data sets, since interactions can be described and quantified. In this paper we study the impact of DSBM as a recon- struction (or filter) technique with respect to source mod- eling. To investigate the performance, different data sets are simulated, different with respect to spatial and tempo- ral characteristics as well as with respect to different noise levels. Since our approach can be viewed as an extension of PCA, source modeling results of PCA-filtered signals are compared with results of DSBM-reconstructed and unfiltered data sets. Dynamical Systems Based Modeling (DSBM) EEG/MEG signals represent electromagnetic poten- tials/fields measured on the scalp surface due to neuronal interactions in the human brain. The high-dimensional complex dynamical system on the neuronal level exhibits low-dimensional behavior on the level of measurements on the scalp surface, observed by correlation dimension studies (Babloyantz et al. 1985) as well as source modeling leading to low-dimensional dipole models. Therefore, one expects that the measurements q(t) (with the vector com- ponents representing the channels of the measurement) can be expressed as a combination of different field maps u i weighted by factors x i depending on time: * Max-Planck-Institute of Cognitive Neuroscience, Leipzig, Ger- many. Accepted for publication: October 24, 2000. Correspondence and reprint requests should be addresssed to Axel Hutt, Max-Planck-Institute of Cognitive Neuroscience, Stephanstr.1a, D-04103 Leipzig, Germany. Fax: ++49 341 9940 221 E-mail: hutt@cns.mpg.de Copyright © 2001 Human Sciences Press, Inc. Brain Topography, Volume 13, Number 3, 2001 219