Two separate source models were created for risk assessment and reward processing applied to 64 EEG-channel data. Regional sources (RS) were seeded for risk assessment according to fMRI activation foci [
6] revealed by the group interaction contrast (high-risk PG > low-risk PG) > (high-risk OG > low-risk OG) and respectively (win PG > lose PG) > (win OG > lose OG) for reward processing as the main focus was on the differences between groups. Applying fMRI constrained source analysis is grounded in a combination of temporal and spatial dynamics while including prior information of fMRI activation foci to improve the spatial validity of the model [
40],[
41]. Source waveforms were calculated using a four-shell spherical head model, which considered characteristics of the brain like, conductance, bone, cerebrospinal fluid, and scalp [
28], and a regularization constant of 1 percent for the inverse operator to reduce the interaction between sources. A RS consists of three equivalent current dipoles with identical location but reciprocally orthogonal orientations [
28],[
42]. As the activity of RS is hardly sensitive to small differences between the modeled location of active brain regions and individual anatomical location [
28],[
43], the obtained source waveforms for the fMRI seeded sources should be quite robust despite including different participants in experiment 1 (fMRI) and experiment 2 (EEG). For risk assessment a multiple source model on group differences (PG vs. OG) of ERP difference waves (high-risk vs. low-risk) was applied. For reward processing the multiple source model was applied on group differences (PG vs. OG) of ERP difference waves (win vs. lose). The fMRI contrast for risk assessment [(high-risk PG > low-risk PG) > (high-risk OG > low-risk OG)] resulted in three activation foci seeded as RS (RS 1: right superior temporal gyrus, RS 2: right inferior frontal gyrus, RS 3: right thalamus; see Additional file
1: Figure S1), which were fixed according to their location in the source model. Furthermore, the fMRI contrast for the reward processing [(win PG > lose PG) > (win OG > lose OG)] yielded three activation foci seeded as RS (RS 1: right superior frontal gyrus, RS 2: left superior parietal lobe, RS 3: left anterior cingulate gyrus; see Additional file
1: Figure S2), which were also fixed according to their location in the source model. To avoid reciprocal interaction, so-called “crosstalk” between sources with a distance of less than 30 mm, they were averaged according to the nearest neighbor method [
27]. This was necessary for the two parietal sources during reward processing, where the coordinates of the two sources were averaged. The new averaged source (see Additional file
1: Figure S2, RS 2) was located within the two centimeter range from its original fMRI peak locations. The averaging of the coordinates of neighboring sources is justified by the integrative nature of RS in a multiple discrete source model. The reason is that errors in the corresponding center location smaller than two centimeters do not sufficiently influence source waveforms, as long as the distances between the different sources are larger [
27],[
40]. Additionally to the three fixed RS, a sequential fitting procedure was applied in BESA for risk assessment to reduce residual variance of the model. The time windows for the phase of risk assessment were defined around the first two peaks in the global field power curve. As there was no additional clear peak in the global field power curve a third long time window (500-940 ms) was defined (see Additional file
1: Figure S1B red boxes). For the 80-160 ms time window two sources, RS 4 and RS 5, were fitted. In the 360-430 ms time window, again two sources, RS 6 and RS 7, were fitted, while RS 4 and RS 5 were switched off. Furthermore, for the 500-940 ms time window two additional sources were fitted, RS 8 and RS 9, while RS 4-7 were switched off. The model for risk assessment, therefore, consisted of 9 RS and explained a variance of 93.4 percent (see Additional file
1: Figure S1). For reward processing, additional to the two fixed RS, a sequential fitting procedure to reduce residual variance was performed. The time windows for the phase of reward processing were defined around the first peak in the global field power. As there was no additional clear peak and a constant high level of global field power a second long time window (200-1000 ms) was defined (see Additional file
1: Figure S2B red boxes). For the 80-160 ms time window, RS 3-5 were additionally fitted, and for the 200-1000 ms time window, RS 6-11 were fitted, while RS 3-5 were switched off. The model for reward processing included 11 RS, which explained 94.2 percent of variance (see Additional file
1: Figure S2).