Fig. 4.

Clinical phenotypes show differences in network architecture. All subplots depicting a statistical comparison significant at a p value of < 0.05 are marked with an asterisk (*). a Covariance matrices of inter-electrode coherence measurements for Classic (n = 26), Hanefeld (n = 4) and PSV (n = 5) groups, allowing visualisation of higher-order network function. Each row and each column represent a pair of electrodes. These are arranged into blocks along the axes, with measures for each electrode pair at all frequency bands and across the overall spectrum. The intensity of each cell represents the covariance between activity in the corresponding electrode pairs at the corresponding frequency band. Visualisation suggests differences in network activities between phenotypes; comparison of first principal components indicated a statistically significant difference between groups (p < 0.0001, Kruskal-Wallis H test), indicating that there are differences in network architecture between clinical phenotypes. The Classic group demonstrates a pattern of low network involvement of left and right sided occipital areas. The Hanefeld network pattern is characterised by very low covariance across all pairs in delta band, including cross-frequency, indicating abnormalities in network function within this frequency range, as well as very low network involvement of right occipital and parietal regions. The PSV groups demonstrates a similar overall pattern to the Classic group, though with greater overall covariance between pairs and a more marked reduction in involvement of bilateral occipital regions. b Subplots of overall covariance matrices in A, showing only covariance between electrode pairs over the whole power spectrum. Each row and each column represents an electrode pair as labelled. Closer examination of covariance within the overall spectrum suggests that differences seen across the whole matrix are still evident within the overall spectrum alone