37. Importance of weak connections in functional network analysis of left TLE

Abstract

Brain can be described as a complex network and its structure can be changed in disease ( Rubinov and Sporns, 2010 ). The analysis based on graph theory is a well established tool for quantification of changes in network topology ( Tijms et al., 2013 , Diessen et al., 2013 ); so far the binary representation of connectivity network has been very popular: it preserves a chosen percentage of the strongest connections, thus simplifies analysis and interpretations of results. This approach, however, discards information about weak connections and weighted representation comes into use. We aim to examine differences in results from both approaches. We used resting state fMRI data from 9 left temporal lobe epilepsy (TLE) patients and 20 controls. We extracted matrices of functional connectivity based on Pearson correlations between representatives from regions defined by AAL atlas ( Tzourio-Mazoyer et al., 2002 ), in case of binary approach we used 15% of the strongest connections. For both representations we computed path length, clustering coefficient, connectivity strength, and betweenness centrality measures and evaluated group differences by Mann–Whitney U test. By weighted approach, we observed the significantly (FDR corr.) higher normalized clustering coefficient and higher normalized characteristic path length in patient group that suggest the shift of the network topology toward more regular structure. Together with increased global betweenness centrality, it leads us to an assumption that the patient network forms densely connected clusters that are mutually connected only by a small number of links, whereas the control network exhibits more distributed and balanced pattern of topology. The most striking difference between groups is in connectivity strength which is significantly lower in left TLE group. These results, however, could not be observed in binary representation of connectivity. The increases in characteristic path length and global clustering coefficient were not significant after FDR correction ( p -values 0.03 and 0.14) and only a trend towards lattice topology can be concluded. Also just a trend to increased betweenness centrality ( p = 0.03) in patient group was seen, no change in connection strength was found. Our findings support other studies showing that network topology in left temporal lobe epilepsy shifts towards regular structure ( Diessen et al., 2013 ). This, combined with changes in betweenness centrality and connection strength, allows us to provide better specification of network topology in left TLE patients. We stress that the difference between groups lies especially in low correlations which is the reason why the analysis based on binary representation failed to find statistically significant changes.