126 Chapter 6 reference electrode standardization technique (REST) following Yao (2001) (26) and Yao e.a. (2019) (27). Recordings were cleaned in EEGLAB and selected plugins using the following steps. First data were filtered, then subjected to an initial independent components analysis (28). IClabel (29) was used to identify eye components and removed. Next, data were cleaned using the artefact subspace reconstruction (ASR) (30). Next, a full ICA was performed and remaining artefactual components identified with ICLabel, removing Eye, Heart, line noise, and muscle artefacts identified with 50% certainty or above. Finally, missing or deleted channels were back-imputed using spherical interpolation. EEG processing Power spectra were obtained using fast Fourier transform (FFT). From these, individual alpha and beta power were derived as average power from 8 to 13 Hz and 14 to 20 Hz respectively. This was done for segments of EEG data of 1s epoch to obtain multiple measures for each individual. Connectivity was established using coherence and weighted PLI (31) between all possible pairs of signals for the same frequency bands (alpha, beta). Global graph metrics -average connectivity, average path length, diameter, rich-club metric, and assortativity- were extracted from the resulting 62 x 62 connectivity matrices based on alpha and beta oscillations using the MITGraphToolbox (32). For a full overview of what these metrics represent we refer to Stam (2014) (21). We use the nondirected weighted network versions of all metrics. Average connectivity reflects the average level of alpha or beta band synchronicity seen between all possible pairs of electrode signals, and may measure the average amount of integration between the distant brain areas. Path length is the graph metric that is inversely related to the level of functional integration of the network (19) and thus the optimality of communication between brain areas for information transfer (21). Diameter is also a related concept, but reflects the maximal distance in the graph. The rich-club metric detects whether there is a community of nodes that preferentially connect to each other, while remaining nodes are precluded from forming a similar club. Rich-club networks are well connected (highly integrated) while maintaining a substantial level of robustness to attacks (i.e. less-of-function when specific brain areas are lost) (22, 33). Assortativity reflects the tendency for nodes to connect to nodes of similar degree. Biological networks often show disassortativity, where high degree nodes preferentially link to low degree ones. Combined, all metrics vary across particular network topologies, specifically, disorganized random networks, rich-club networks, and hierarchically organized scale-free networks. Statistical analyses The data were analysed with statistical techniques accommodating the nature of the data (multiple measures per person for EEG power; single measure per person for network and graph measures) and the different comparisons (within-subject / repeated measures over time / between subjects). Within-subject comparisons were performed only for the power data in 1s
RkJQdWJsaXNoZXIy MjY0ODMw