Edge-centric analysis of time-varying functional brain networks with applications in autism spectrum disorder

Abstract

The interaction between brain regions changes over time, which can be characterized using time-varying functional connectivity (tvFC). The common approach to estimate tvFC uses sliding windows and offers limited temporal resolution. An alternative method is to use the recently proposed edge-centric approach, which enables the tracking of moment-to-moment changes in co-fluctuation patterns between pairs of brain regions. Here, we first examined the dynamic features of edge time series and compared them to those in the sliding window tvFC (sw-tvFC). Then, we used edge time series to compare subjects with autism spectrum disorder (ASD) and healthy controls (CN). Our results indicate that relative to sw-tvFC, edge time series captured rapid and bursty network-level fluctuations that synchronize across subjects during movie-watching. The results from the second part of the study suggested that the magnitude of peak amplitude in the collective co-fluctuations of brain regions (estimated as root sum square (RSS) of edge time series) is similar in CN and ASD. However, the trough-to-trough duration in RSS signal is greater in ASD, compared to CN. Furthermore, an edge-wise comparison of high-amplitude co-fluctuations showed that the within-network edges exhibited greater magnitude fluctuations in CN. Our findings suggest that high-amplitude co-fluctuations captured by edge time series provide details about the disruption of functional brain dynamics that could potentially be used in developing new biomarkers of mental disorders.

The diversity and multiplexity of edge communities within and between brain systems

“…We calculated eFC by first Z-scoring each regional time series and computing the element-wise product between all pairs of time series, yielding M = 79,800 unique pairs corresponding to every possible edge. Using these so-called edge time series (Zamani Esfahlani et al, 2020b, 2021aBetzel et al, 2021;Greenwell et al, 2021;Faskowitz et al, 2020), we calculated the 79,800 3 79,800 eFC matrix of all pairwise similarities (see STAR Methods for details). This procedure was repeated separately for each of the 10 subjects in the MSC and for each of their 10 scans.…”

“…Similarly, eFC can be used to detect pervasively overlapping communities, yielding new insight into the brain's modular structure (Faskowitz et al, 2020). Although recent work suggests that these and other features of edge-centric analyses can be exploited to learn more about brain organization and dynamics, few studies have systematically compared them with more common methods (Zamani Esfahlani et al, 2021a;Novelli and Razi, 2021). Future work should investigate these questions in greater detail.…”

Hierarchical organization of spontaneous co-fluctuations in densely-sampled individuals using fMRI

“…Although events occur briefly and infrequently, the pattern of whole-brain co-fluctuations expressed during these periods necessarily contribute more to the time-averaged FC than lower-amplitude frames [3]. Moreover, high-amplitude co-fluctuation patterns can be partitioned into a small number of recurring clusters or “states” [6–8], encode information about subjects’ brainbased fingerprints [9], and can possibly enhance brain-behavior correlations [3].…”

BOLD cofluctuation ‘events’ are predicted from static functional connectivity

“…Esfahlani and colleagues found that "events," time points with the highest BOLD signal cofluctuation, reproduce static functional connectivity patterns better than the same number of "non-events," time points with the lowest BOLD signal cofluctuation, and require relatively few timepoints to reproduce them well. The authors concluded that rather than functional network structure being present at all timepoints, it is driven by events -a discrete and temporally sparse phenomena (Esfahlani et al, 2020).This idea has deep implications for the field: a thorough analysis of events across brain organizational levels (e.g., from systems to cellular recordings) could reveal information about the physiological mechanisms of FC and new analysis methods focused on events could improve the clinical utility of fMRI (Esfahlani et al, 2021;Greenwell et al, 2021). The idea that brain network information can be identified in a reduced data set is not new.…”