2.5. Connection Definition and Threshold Selection

For the connection definition and threshold selection, we followed the methods of one of our previous researches [31]. The Pearson correlation coefficient was used to define the network edge in this study. First, we calculated the mean time courses of each node and then performed multiple linear regression to remove the pseudodifferences caused by head movement. The residuals were used to compute the partial correlation, producing an N × N correlation matrix, where N represents the number of nodes in a given parcellation. According to the predetermined threshold, the correlation matrix was converted into a binary matrix (For the mathematical definition of the Pearson correlation coefficient, see Supplemental Material ).

Sparsity, S, which is the ratio of the number of real existing edges to the maximum possible number of existing edges [35], was used as the threshold setting. This method has been widely adopted in similar studies [36–39]. To ensure the comparability of results between the parcellations, the threshold space S∈(8%, 32%) of the AAL90 parcellation was used as the standard for all five parcellations, and the brain functional networks of all subjects were constructed with an S step size of 5% within the threshold space (For details of the threshold selection criteria, see Supplementary Material ).

To characterize the integrity properties of a metric in the complete sparsity space, we calculated the area under the curve (AUC) for each metric. AUC provides a method to assess the total change in the network node properties under different degrees of sparseness. This method has been applied in previous research, which showed that it represents a very sensitive method to assess changes in the topological properties of a brain network [21].