Graph analysis of nonlinear fMRI connectivity dynamics reveals distinct brain network configurations for integrative and segregated information processing

Abstract

The human brain is organized into functional networks, whose spatial layout can be described with functional magnetic resonance imaging (fMRI). Interactions among these networks are highly dynamic and nonlinear, and evidence suggests that distinct functional network configurations interact on different levels of complexity. To gain new insights into topological properties of constellations interacting on different levels of complexity, we analyze a resting state fMRI dataset from the human connectome project. We first measure the complexity of correlational time series among resting state networks, obtained from sliding window analysis, by calculating their sample entropy. We then use graph analysis to create two functional representations of the network: A ‘high complexity network’ (HCN), whose inter-node interactions display irregular fast changes, and a ‘low complexity network’ (LCN), whose interactions are more self-similar and change more slowly in time. Graph analysis shows that the HCNs structure is significantly more globally efficient, compared to the LCNs, indicative of an architecture that allows for more integrative information processing. The LCNs layout displays significantly higher modularity than the HCNs, indicative of an architecture lending itself to segregated information processing. In the HCN, subcortical thalamic and basal ganglia networks display global hub properties, whereas cortical networks act as connector hubs in the LCN. These results can be replicated in a split sample dataset. Our findings show that investigating nonlinear properties of resting state dynamics offers new insights regarding the relative importance of specific brain regions to the two fundamental requirements for healthy brain functioning, that is, integration and segregation.

Diverging functional connectivity timescales: Capturing distinct aspects of cognitive performance in early psychosis

“…For the LEN, highest DC values belonged to pre-and postcentral SM and DAT regions (bilaterally), as well as to bilateral prefrontal-and cingulum areas of the CC (Figure 6). Overall, the DC values were spatially organized in strong correspondence with our previous results in large sample of young and healthy subjects (Hirsch & Wohlschlaeger, 2022). 2022) for details (Figure 8).…”

“…Moreover, we performed basic topological analyses at the node-level, based on binarized versions of the (group-level) HEN and LEN templates, derived from the controls. This was done to replicate our previous finding that ESE is topographically organized along a subcortical (SC) to cortical axis in healthy subjects (Hirsch & Wohlschlaeger, 2022). Finally, to gam more insight into the neurobiological mechanisms underlying HEN and LEN configurations at the cortical level, we analyze their spatial correspondence with neurotransmitter maps derived from positron emission tomography (PET) and the topography of brain rhythms from magnetoencephalography (MEG).…”

“…One SampEn value was then computed for each correlational timeseries, resulting in a vector with 6670 elements for each subject. In accordance with previous work, we then constructed high-entropy (HEN) and low-entropy (LEN) network-templates, by selecting edges with the highest and lowest mean ESE values across healthy subjects (Hirsch & Wohlschlaeger, 2022). This was done for a range of different thresholds, and for every threshold we extracted the corresponding ESE values from the patients, which were then correlated across patients to derive the similarity matrices to be put into the model (Ge et al, 2016).…”

“…Nevertheless, the observed relationships between ESE at different levels and behavior, oscillatory dynamics and neurotransmitter densities suggest relevant information can be extracted from the timescales of TVFC. Finally, in the past we and others have equated high (single-scale) SampEn with high complexity (Hirsch & Wohlschlaeger, 2022;Jia & Gu, 2019a, 2019bJia et al, 2017), but others have argued that such an interpretation requires a multi-scale entropy analysis (Costa, Goldberger, & Peng, 2002;A. C. Yang et al, 2015).…”

“…Although TVFC is still a controversial topic (Liegeois, Laumann, Snyder, Zhou, & Yeo, 2017; Lurie et al, 2020), studying it has proven to be informative regarding interindividual differences (Liegeois et al, 2019; Vidaurre, Llera, Smith, & Woolrich, 2021) and disease states (Jia, Gu, & Luo, 2017; Kaiser et al, 2016; Sakoglu et al, 2010). Consequently, a small number of resting-state fMRI studies have used sample entropy (SampEn) (Richman & Moorman, 2000) to quantify the self-similarity of edge fluctuations (edge-SampEn [ESE]) derived from sliding-window analysis (Hirsch & Wohlschlaeger, 2022; Jia & Gu, 2019b; Jia et al, 2017). SampEn is one way of assessing INT, with higher values corresponding to shorter INT (Omidvarnia, Mesbah, Pedersen, & Jackson, 2018; Sokunbi et al, 2014), and it is also significantly associated with mental abilities and cognitive load in healthy subjects (S. S. Menon & Krishnamurthy, 2019; Nezafati, Temmar, & Keilholz, 2020; Omidvarnia et al, 2022; Omidvarnia et al, 2021).…”