To avoid artifacts, peak-to-peak differences of more than 3.5 pT/cm resulted in the rejection of an epoch. After artifact rejection, on average more than 90 valid trials Selleckchem Quizartinib per session remained for event-related averaging. As the amplitude of MEG waveforms was strongly dependent on the individual’s head size and the head position in the MEG device, we did not use the sensor data for analysis. Instead,

we employed distributed source modeling in an empirical Bayesian approach, as implemented in SPM8 (Wellcome Trust Centre for Neuroimaging, University College, London, UK), to reconstruct the cortical sources generating the magnetic-evoked field in response to omission. Subjects’ individual anatomical magnetic resonance images were spatially normalised to a Montreal Neurological Institute (MNI) template brain. The inverse of this

spatial transformation parameter was used to warp a cortical template mesh to the individual magnetic resonance space. The co-registration between MEG sensor positions and the head magnetic resonance imaging was achieved by manually detecting three fiducial points (nasion and the left and right pre-auricular) in the magnetic resonance image that were defined by magnetic resonance markers and the head shape that was measured using a spatial digitiser. To generate the forward model, the lead-field for each sensor was calculated for dipoles at Tanespimycin in vivo each point in the canonical cortical mesh (8196 vertices) by using a single shell model and the ‘forwinv’ toolbox, which SPM shares with Fieldtrip (Oostenveld et al., 2011). The model was then inverted using restricted maximum likelihood

and the multiple sparse priors algorithm (Phillips et al., 2005; Mattout et al., 2006; Friston et al., 2008) for each session separately. In each session, in order to reduce inter-individual variances, each subject’s smoothed images were automatically normalised by SPM using the mean of the entire time period. Because we were mainly interested in the cortical distribution of the omission-related response, which was found in the time window of 100–200 ms after the not omission onset in the previous studies (Yabe et al., 1998; Rüsseler et al., 2001; Bendixen et al., 2009; Horváth et al., 2010; Todorovic et al., 2011; Wacongne et al., 2011), the reconstructions were averaged in the time window of 100–200 ms and the mean reconstruction maps were exported as three-dimensional voxel-based images into MNI space. Finally, the images were smoothed using a Gaussian filter with 8 mm full width half maximum and used for group analysis. For the group analysis, general linear model-based statistical analysis using random field theory was conducted using SPM8.