Fig. 5.1
Spatial maps of 48 intrinsic connectivity networks (a) and the stationary functional connectivity (similarity S matrix); (b) between them in healthy control subjects (HC) and schizophrenia patients (SZ) (Reprint with permission from Yu et al. 2015). Intrinsic connectivity networks are divided into groups and arranged based on their anatomical and functional properties. Functional connectivity was averaged over all subjects in each group. AUD auditory, CB cerebellar, CC cognitive control, DM default mode, SM somatomotor, VIS visual
A dynamic connectivity analysis of each participant was completed by combining a sliding time window approach (with a width of L = 20 repetition time (TR) in 1-TR steps) with a graph theory approach. A sliding time window approach was used to obtain dynamic time courses so that ICN correlations could be computed during each dynamic time course. Accordingly, the authors were able to examine all d-FNCs for each subject. Graph theory was then applied to compute connectivity strengths, clustering coefficients, and global efficiencies for d-FNCs using the brain connectivity toolbox (http://www.brainconnectivity-toolbox.net/; BCT). To determine relationships among d-FNCs, the authors applied first- and second-level analyses: first, the modularity algorithm of Newman (2006) was applied to the window correlation metrics of each subject, averaging the d-FNC within the same module, to yield a total of 554 states (276 states in HCs and 278 states in SZs). Second, the algorithm was applied to all 554 states to produce one final state (Fig. 5.2).
![A417075_1_En_5_Fig2_HTML.gif](https://i0.wp.com/neupsykey.com/wp-content/uploads/2018/01/A417075_1_En_5_Fig2_HTML.gif?w=960)
![A417075_1_En_5_Fig2_HTML.gif](https://i0.wp.com/neupsykey.com/wp-content/uploads/2018/01/A417075_1_En_5_Fig2_HTML.gif?w=960)
Fig. 5.2
Flowchart of the algorithmic pipeline for the first-level connectivity state analysis (Reprint with permission from Yu et al. 2015)
Of note, only one connectivity state is identified by the second-level analysis. In this example, a total of 271 first-level connectivity states (155 states in 75 HCs and 116 states in 67 SZs) that were highly associated with one another were averaged into the second-level state. Compared with HCs, SZs exhibited fewer first-level connectivity states that were subsequently associated with the second-level connectivity state. Moreover, the 155 first-level states in HCs showed higher (P < 0.01, two-sample t-tests of the means and permutation tests of medians) graph metrics than the 116 first-level states in SZs (Fig. 5.3). It is noteworthy that, visually, the pattern of the second-level connectivity state in Fig. 5.3 resembles the stationary connectivity pattern shown in Fig. 5.1.
![A417075_1_En_5_Fig3_HTML.gif](https://i0.wp.com/neupsykey.com/wp-content/uploads/2018/01/A417075_1_En_5_Fig3_HTML.gif?w=960)
![A417075_1_En_5_Fig3_HTML.gif](https://i0.wp.com/neupsykey.com/wp-content/uploads/2018/01/A417075_1_En_5_Fig3_HTML.gif?w=960)
Fig. 5.3
Schematic of the connectivity patterns (Reprint with permission from Yu et al. 2015). (a) node size represents nodal connectivity strength (edge threshold = 0.65); (b) structures and distribution of graph metrics (c) (mean and bootstrapped 95% confidence intervals are in red; box plots and smoothed density histograms are also shown) for the first-level connectivity states related to the second-level connectivity state in healthy control subjects (HC, 155 states) and schizophrenia patients (SZ, 116 states), respectively
The above work reveals a new way to discover and explore the dynamic properties of network functional connectivity in the context of the healthy brain and neurological disease. An improved understanding of how these properties are altered in various disease states will improve the utility of dynamic connectivity research for understanding disease pathology and therapeutic opportunities in neurological research.
5.2.3.2 A Dynamic Variety Analysis of Specific Brain Region Functional Connectivity
In the above study, we learned that a dynamic analysis approach can be used to identify neural biomarkers of mental illness. Next, we will introduce a different approach that was used by Nomi and colleagues to analyze the dynamic properties of a specific brain region, the insula (Nomi et al. 2016).
Nomi and colleagues performed an analysis on resting-state fMRI scan data from 31 healthy adult subjects (ages 18–40 years) downloaded from the Nathan Kline Institute database. Data preprocessing, group ICA, and post-processing were performed in a manner similar to that in the previously discussed study by Yu et al. (Fig. 5.4); however, this work used the DPARSF-A toolbox (http://rfmri.org/DPARSF) for data preprocessing, selected 52 no-noise ICs, and identified four of them as belonging to the insular cortex (the dorsal, ventral, posterior, and middle insula).
![A417075_1_En_5_Fig4_HTML.gif](https://i0.wp.com/neupsykey.com/wp-content/uploads/2018/01/A417075_1_En_5_Fig4_HTML.gif?w=960)
![A417075_1_En_5_Fig4_HTML.gif](https://i0.wp.com/neupsykey.com/wp-content/uploads/2018/01/A417075_1_En_5_Fig4_HTML.gif?w=960)
Fig. 5.4
Schematic of the dynamic variety analysis preprocessing performed by Nomi et al. (Reprint with permission from Nomi et al. 2016). (a) A high-model group independent component analysis (ICA; 100 components) was used to create a functional parcellation of the brain, resulting in 52 non-noise components. (b) Subject-specific time courses from the group ICA were then used to compute functional connections. The static analysis entailed computing correlations over the whole duration of the resting state. The dynamic analysis included acquiring correlation matrices of each subject by employing 45-s tapered-sliding windows (in 1-TR steps) and extracting the connections between each insular subdivision and all other ICS. (c) A concatenated data matrix from step B received k-means clustering using values 2–20 that identified the optimal k as 5 using the elbow criterion; the value of k = 5 then assigned each window to dynamic state k regardless of subject assignment. Finally subject-specific medians for each state k were calculated and averaged together to produce a total of five final dynamic insular states
S-FNCs were computed for each subject, and one-sample t-tests were conducted to identify significant positive connectivities between insular subdivisions and other ICs. In order to assess differences in connectivity strength between insular subdivisions and other ICs, t-tests were conducted on the functional connectivity values. Dynamic functional connectivity was computed in the same way as that described in Yu et al.; however, instead of using graph theory to cluster all d-FNCs, the K-means algorithm was used to partition data into a set of separate clusters, producing final five dynamic insula connectivity states. State 3 was the most frequent occurring insular state (38%; n = 31) and was analogous to the s-FNC finding in that unique functional profiles for each insular subdivision could be observed, but State 3 was much smaller in magnitude than s-FNC. State 1 (24%; n = 31), State 4 (13%; n = 26), and State 5 (20%; n = 30) were moderately represented. Finally, State 2 (5%; n = 59) was the most infrequent insular state. One-sample t-tests were used to evaluate insular connectivity states in a manner similar to that used for s-FNC data.
Figure 5.5 summarizes the significant positive connections found in the s-FNC analysis. In the static condition, the positive connections between insular subdivisio ns and other ICs displayed as follows: the connections between the dorsal anterior insula and frontal areas, the connections of the middle and posterior insula with sensorimotor areas, and the connections between ventral insula and ICs representing affective subcortical areas including the nucleus accumbens, hippocampus, and amygdala. These connections are consistent with the emotion-cognition-interoception framework of insular subdivision function (Cauda et al. 2011).
![A417075_1_En_5_Fig5_HTML.gif](https://i0.wp.com/neupsykey.com/wp-content/uploads/2018/01/A417075_1_En_5_Fig5_HTML.gif?w=960)
![A417075_1_En_5_Fig5_HTML.gif](https://i0.wp.com/neupsykey.com/wp-content/uploads/2018/01/A417075_1_En_5_Fig5_HTML.gif?w=960)
Fig. 5.5
Plots of significant positive connections between insular subdivisions and other ICs (Reprint with permission from Nomi et al. 2016). CB cerebellum, CEN central executive network, DMN default mode network
Also depicted in Fig. 5.5 are positive correlations between insular subdivisions and other ICs in each dynamic state. State 3 was similar to the s-FNC. In State 1, all four insular subdivisions showed connections with sensorimotor, temporal, visual, central executive network, and salience network ICs. In addition, the middle insula also exhibited connections with the cerebellum, while the posterior insula was the only region that did not have connections with frontal areas. The results suggest that all the insula subdivisions keep in step with the frontal cortex to process and integrate sensorimotor and visual information in State 1. In State 2, only the dorsal anterior insula appeared connections with subcortical ICs and DMN, while the medial, posterior, and ventral insula showed connectivity with temporal, sensorimotor, and salience network ICs. This suggests that in State 2, the medial, posterior, and ventral subdivisions of the insula work together to coordinate sensorimotor and temporal information, while the dorsal insula works independently to coordinate communication between subcortical areas and the DMN.
Figure 5.6 shows a polar plot of s-FNC correlations for the four subdivisions of the insula. Significant differences in connection strengths between each subdivision with other ICs are identified by an asterisk placed along the radiating axis. The dorsal anterior insula showed stronger connections with frontal brain areas than all other insula subdivisions, particularly with the frontal pole. By contrast, the ventral insula showed significantly stronger connections with ICs representing the nucleus accumbens, hippocampus, and amygdala. These findings are in accordance with the findings of previous studies (Cauda et al. 2011, 2012).
![A417075_1_En_5_Fig6_HTML.gif](https://i0.wp.com/neupsykey.com/wp-content/uploads/2018/01/A417075_1_En_5_Fig6_HTML.gif?w=960)
![A417075_1_En_5_Fig6_HTML.gif](https://i0.wp.com/neupsykey.com/wp-content/uploads/2018/01/A417075_1_En_5_Fig6_HTML.gif?w=960)
Fig. 5.6
Polar plot displaying functional connections between insular subdivisions and ICs in the static functional connectivity analysis (Reprint with permission from Nomi et al. 2016). dAI dorsal anterior insula, STG superior temporal gyrus, MTG medial temporal gyrus, IFG inferior frontal gyrus, CG central gyrus, SMA supplementary motor area, OCC occipital, CC calcarine cortex, FG fusiform gyrus, ACC anterior cingulate cortex, OBF orbitofrontal cortex, DLPFC dorsal lateral prefrontal cortex, AG angular gyrus, MPFC medial prefrontal cortex, DMN default mode network, SMG supramarginal gyrus
Figure 5.7 shows a similar polar plot to that in Fig. 5.6, but instead represents functional connections in each of the five dynamic states. State 1 had 41 significant differences across the insular subdivisions, State 2 had 33, State 3 had 77, State 4 had 61, and State 5 had 113. States with fewer significant changes accordingly represent that insular subdivisions exhibited more common connections to various ICs (convergence across subdivision connections), while states with more changes mean that insular subdivisions exhibited more individual connectivity profiles with various ICs (divergence across connections).