Meditation-induced effects on whole-brain structural and effective connectivity

Fig. 1. Model parameters estimation (adapted from Gilson et al. 2020). To capture BOLD dynamics, the model uses the time-series parcellated into 116 ROIs (box on the left top) to calculate two autocovariance matrices (FC0 and FC1 matrices) both with and without time lag (blue and green lines in the central box on top). The SC connectivity matrix (black and white matrix on the left) is used as a binary matrix to constrain the model’s topology to existing connections. Besides the effective connectivity estimates (Pink matrix at the left bottom), the autocovariance matrices (green and blue matrices at the right bottom) are also reproduced by the dynamic model. The model undergoes an optimization procedure so that, at each step, the estimated FC matrices are evaluated with regards to the empirical FC0 and FC1 matrices. The optimization steps are repeated until reaching a high Pearson correlation coefficient between the model’s and the empirical FC matrices, reducing this way the model error (depicted as the green and black lines in the central box on the left of the figure)

Fig. 2. Intersected SC matrix for each group. To estimate functional dynamics of both groups during each condition, we intersected the SC matrix specific to each group (two matrices on the left side) and we generated a new SC matrix with all the links above-threshold common to both meditators and controls (SC matrix on the right side)

Fig. 3. Resting-state vs. meditation: classification accuracies within each group. Accuracy distributions are shown for meditators (left panel) and controls (right panel) The violin plots in green show the accuracies for the MLR classifier using the features related to EC, while the blue ones represent MLR performance using FC features. Accuracies related to 1NN are shown in orange for EC and in pink for FC features. For both groups, all the combinations between classifiers (MLR and 1NN) and metrics were significantly higher than chance-level (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p<0.0001$$\end{document}p<0.0001)

Fig. 4. Results of group classification during meditation and resting-state. The performance of 1NN and MLR in discriminating the two groups are shown for resting-state condition (A) and meditation task (B). Significant differences in performance depending on the type of features are shown in black for the contrast EC-FC features and in purple for the contrast EC-SC features. Violin plots colored in gray represent accuracy distributions of surrogate data. EC features led to the highest average accuracies with either the MLR or the 1NN, depending on the condition. While the 1NN yielded the best performance for the resting-state sessions (average accuracy of 67%), the MLR outperformed the 1NN when comparing the two groups during the meditation task (average accuracy of 68% against 55%, respectively)

Fig. 5. RFE results: support network for group classification during meditation. The group signatures (i.e., highest-ranked EC features) are presented for the two conditions in sagittal, medial, and dorsal views, both for the right and left hemispheres. The color of the nodes is determined by their belongingness to one of the eight resting-state networks (red = visual; orange = somatomotor; light orange = dorsal attention; light green = salience/ventral attention; dark green = limbic; light blue = control; blue = DMN; dark blue = subcortical). The networks are drawn from the parcellation in 100 ROIs comprising 7 functional divisions (Schaefer et al. 2018) together with 16 subcortical regions (Tian et al. 2020). Gray arrows show the directionality of connection (EC-links) between ROIs of distinct networks. In contrast, direct links between nodes belonging to the same network are colored according to network-specific hue. Visualization of results was generated using the BrainNet Viewer Toolbox (Xia et al. 2013)

Fig. 6. Increased network-related SC in the meditator group compared to the control group. A The significant network resulting from the contrast meditators > controls, comprising three edges and four nodes, is represented from a lateral view of the left hemisphere. Nodes are colored according to the hue of the network they belong to as displayed in the legend on the left side. B Visualization of the three pathways found to have increased connectivity in the meditator group. ROIs are colored as follows: blue for left putamen, pink for left somatomotor, red for frontal eye field (FEF), and green for left visual area. Pathways are colored according to the number of participants that have a tract passing through the voxel, with yellow representing the heighest value and black the smallest, as represented in the legend on the right side of the figure