Fig. 1.

Methods. A) Experimental design: Immediately before (PRE) and after (POST) NFB training, resting-state and diffusion-weighted imaging (DWI) data were acquired. The training started with a motor execution task in which a 2-class support vector machine (SVM) algorithm was trained to discriminate between motor execution and rest based on the distributed voxel patterns. The NFB (or sham) training included 3 runs, in which participants performed a motor imagery task with the aid of the NFB (either real or sham) to recruit brain areas associated with motor execution obtained previously. B) MOU-EC model workflow (adapted from Gilson et al. 2019): The estimation of model parameters is done using the resting-state functional data parcellated into 116 ROIs and the SC connectivity matrix (black and white matrix on the left), with the latter used as a binary matrix to constrain the model’s topology. From the resting-state timeseries the empirical BOLD autocovariance matrices are calculated (blue and green matrices on the right of top panel) both with and without time lag (FC1 and FC0, respectively), and then reproduced by the model. The model’s estimation of the autocovariance matrices and effective connectivity matrix undergoes the Lyapunov optimization procedure using a gradient descent algorithm which minimizes the model error between the model’s estimate and the empirical FC0 and FC1.The iterative optimization procedure is repeated until the best fit between the model’s estimate and the empirical FC matrices is reached (high Pearson correlation coefficient), thus reducing the model error to the minimum (green and black lines in the left upper panel of Fig. 1B. C) Algorithm to calculate the intrinsic ignition. Adapted fromDeco and Kringelbach (2017): The process to calculate the intrisic ignition involve the following steps: -Spiking neurons (green dot) produce driving events, which are captured applying a threshold as explained in Tagliazucchi et al. (2012). - For each driving event (gray area), the activity in the rest of the network is calculated within the time-window of 4TR (the whole red area). -A binarized matrix (black-and-white matrix on the right) is constructed for each time window according to the reoccurrence of driving events from different brain areas. -The largest subcomponent (blue and red graphs at the bottom) of the binary matrix represents a measure of global integration. -The previous steps are repeated for each driving event, thus allowing to calculate the mean and the variability of the Intrinsic-Driven integration of each brain area.