Submission No: 1743 Authors: G Strobbe1, J Lopez2, P van Mierlo1, M de Vos3, B Mijovic4, S Van Huffel4, H Hallez5, S Vandenberghe1 Institutions: 1Ghent University IBBT, Departement of Electronics and Information Systems, MEDISIP, Ghent, Belgium, 2National University of Colombia, Bogotá, Colombia, 3KU Leuven, Department of Electrical Engineering-ESAT, SCD-SISTA and IBBT Future Health Department, Leuven, Belgium, 4KU Leuven, Department of Electrical Engineering-ESAT, SCD-SISTA and IBBT Future Health Department, Leuven, Belgium, 5Catholic University College of Bruges-Ostend, Ostend, Belgium Introduction: A parametric empirical Bayesian (PEB) framework for distributed sources is recently introduced in the widely used neuroimaging package SPM (Henson, 2011). The framework offers multi-modal and multi-subject integration and the ability to test forward models using real data in Bayesian model comparison based on free energy. Within the PEB approach, the multiple sparse priors (MSP) algorithm is the state-of-the-art technique in which multiple cortical sources with compact spatial support, specified in terms of empirical priors, are automatically selected (Friston, 2008). More realistic forward modeling can be introduced in the MSP approach to further increase the precision of source estimation. In the current implementation, the source-space is divided into a number of small cortical patches calculated based on dipoles distributed on a cortical mesh. The field propagation of the surface patches is calculated based on solutions for a 3-layered, scalpskull-brain Boundary Element Method (BEM) approximation to the head, see figure 1A. In reality, dipoles can also be located inside gray matter. More realistic volume-conductor models including gray matter can be modeled based on a high resolution anatomical MR image and finite difference method (FDM) forward modeling (Hallez, 2007), see figure 1B. As such and in analogy with the MSP algorithm based on cortical patches, the source-space can be divided into a number of small volumetric regions in gray matter. Methods: A Finite Difference Method forward solver based on reciprocity (van Rumste, 2000) was introduced in the parametric empirical Bayesian framework of SPM. This allowed us to model dipoles inside gray matter and therefore we could generalize the surface patch generation process to the construction of multiple volumetric sparse regions. We used volume-conductor models based on the MNI template which is used by default in SPM. A 2D cortical mesh was constructed including 7006 dipoles to use with BEM forward modeling, see figure 1A. This mesh was extended to dipoles located inside gray matter with an inter-dipole distance of 2mm, based on a head model including gray matter and CSF, see figure 1C. This new volumetric forward modeling approach was compared with the standard MSP forward modeling in terms of free energy and the plausibility of source reconstructions with previously published real data (De Vos, 2012). In brief, twenty healthy individuals (12 females, age 20–28 years) participated in a visual detection paradigm in which face, house, inverted face and words stimuli were presented. We used 83 channels grand averaged datasets over subjects for each task.