Figure 2
Resting-state effective connectivity associated with neurofeedback training. (A) A fully connected dmPFC and bilateral amygdala network architecture that was used to model progressive changes (over training runs) in resting-state effective connectivity. The reciprocal connections between ROIs and their self-connections are indicated with arrows. (B) The average effect of neurofeedback training across resting-state runs for commonalities (i.e., the mean connectivity across both groups) and differences between experimental and control groups. Note that DCM parameters per group can be derived using average and difference equations (panel F). (C) The progressive effect for commonalities and differences between experimental and control groups. Progressive self-inhibition of dmPFC was significantly lower in the experimental group. Negative values are depicted in blue, positive in red. Connectivity strength is illustrated by the proportional arrow thickness with posterior probability (Pp) in brackets. Significant connectivity parameters are indicated with shaded rectangles and bold font. Asterisks denote statistical significance (** for Pp > .99, * for Pp > 0.95). (D) Network architecture of target models. Fully-connected, top-down and bottom-up DCM models without direct connections between bilateral amygdala were selected to find the most likely resting-state network model associated with learning control of top-down self-regulation over bottom-up self-regulation (Fig. 1C). (E) Posterior probability of models for commonalities and differences across experimental and control groups. The top-down model explained average and progressive effects remarkably better than other models and was more evident in the experimental than in the control group. (F) An illustration of the significantly lower progressive self-inhibition of dmPFC in the experimental as compared to the control group, as indicated by lower positive self-connectivity values given intrinsically negative self-connectivity parameters.
