A Method for Estimating Dynamic Functional Network Connectivity Gradients (dFNGs) From ICA Captures Smooth Inter-Network Modulation

Abstract

Dynamic functional network connectivity (dFNC) analysis is a widely used approach for studying brain function and offering insight into how brain networks evolve over time. Typically, dFNC studies utilize fixed spatial maps and evaluate transient changes in coupling among time courses estimated from independent component analysis (ICA). This manuscript presents a complementary approach that relaxes this assumption by spatially reordering the components dynamically at each time point to optimize for a smooth gradient in the FNC (i.e., a smooth gradient among ICA connectivity values). Several methods are presented to summarize dynamic FNC gradients (dFNGs) over time, starting with static FNC gradients (sFNGs), then exploring the reordering properties as well as the dynamics of the gradients themselves. We then apply this approach to a dataset of schizophrenia (SZ) patients and healthy controls (HCs). Functional dysconnectivity between different brain regions has been reported in SZ, yet the neural mechanisms behind it remain elusive. Using resting-state fMRI and ICA on a dataset consisting of 151 SZ patients and 160 age and gender-matched HCs, we extracted 53 intrinsic connectivity networks (ICNs) for each subject using a fully automated spatially constrained ICA approach. We develop several summaries of our functional network connectivity gradient analysis, both in a static sense, computed as the Pearson correlation coefficient between full time series, and a dynamic sense, computed using a sliding window approach followed by reordering based on the computed gradient, and evaluate group differences. Static connectivity analysis revealed significantly stronger connectivity between subcortical (SC), auditory (AUD), and visual (VIS) networks in patients, as well as hypoconnectivity in the sensorimotor (SM) network relative to controls. sFNG analysis highlighted distinctive clustering patterns in patients and HCs along cognitive control (CC)/default mode network (DMN), as well as SC/AUD/SM/cerebellar (CB) and VIS gradients. Furthermore, we observed significant differences in the sFNGs between groups in SC and CB domains. dFNG analysis suggested that SZ patients spend significantly more time in a SC/CB state based on the first gradient, while HCs favor the SM/DMN state. For the second gradient, however, patients exhibited significantly higher activity in CB domains, contrasting with HCs' DMN engagement. The gradient synchrony analysis conveyed more shifts between SM/SC networks and transmodal CC/DMN networks in patients. In addition, the dFNG coupling revealed distinct connectivity patterns between SC, SM, and CB domains in SZ patients compared to HCs. To recap, our results advance our understanding of brain network modulation by examining smooth connectivity trajectories. This provides a more complete spatiotemporal summary of the data, contributing to the growing body of current literature regarding the functional dysconnectivity in SZ patients. By employing dFNG, we highlight a new perspective to capture large-scale fluctuations across the brain while maintaining the convenience of brain networks and low-dimensional summary measures.

Keywords: dynamic functional network connectivity (dFNC); dynamic functional network connectivity gradient (dFNG); gradient; independent component analysis (ICA); schizophrenia.

FIGURE 1

Composite maps of the 53 identified intrinsic connectivity networks (ICNs), divided into seven functional domains: subcortical (SC), auditory (AUD), sensorimotor (SM), visual (VIS), cognitive control (CC), default mode network (DMN), and cerebellar (CB) network.

FIGURE 2

Schematic depicting the proposed method. The fMRI data were preprocessed using standard procedures, and then spatially constrained ICA was run on the data using the Neuromark fMRI 1.0 template, resulting in 53 ICNs. Next, FNC was calculated using a sliding window approach. A diffusion map (gradients) was computed for each windowed‐dFNC. Each dFNC matrix was reordered based on its gradient, followed by k‐means clustering of the reordered dFNC. This resulted in five dFNGs.

FIGURE 3

The average of (a) original static FNC, (b) reordered average FNC based on gradient 1, and (c) reordered average FNC based on gradient 2 for schizophrenia group.

FIGURE 4

Visualization of the interaction between the average of the first two cortical gradients for (a) patients (SZ) and (b) HC groups in 2D view. The three dotted patterns correspond to VIS (green), SC/AUD/SM/CC/DMN/CB (blue), and CC/DMN (red) networks for HC and VIS (green), SC/AUD/SM/CB (blue), and CC/DMN (red) networks for SZ.

FIGURE 5

(a) The reordered FNC based on the first gradient associated with SZ patients, (b) the group differences between SZ patients and HCs defined as $-10{\log }_{10}\left(\text{pvalue}\right)\times \mathrm{sign}\left(\text{tvalue}\right)$, and (c) the spatial map associated with the difference between HC and SZ, generated by multiplying the normalized component maps with the sign‐corrected gradients for each individual, which were then summed across components to obtain a subject‐specific representation of functional connectivity. Finally, a voxel‐wise t‐test was applied to compare the spatial maps between healthy controls and schizophrenia patients. Regarding the sFNG analysis, the SZ group showed hypoconnectivity in subcortical (SC) and cerebellar (CB) domains.

FIGURE 6

Schematic depicting the state transition (cluster centroids) for (a) original dFNC, (b) dFNG based on gradient #1, and (c) dFNG based on gradient #2.

FIGURE 7

The 3D spatial maps associated with each state based on gradient #1. A spatial map of the dFNGs was created by thresholding and normalizing each component map, followed by using the normalized cluster centroids obtained from gradient #1 k‐means clustering as the weight for the component maps to create spatial maps.

FIGURE 8

The 3D spatial maps associated with each state based on gradient #2. A spatial map for dFNG was created by thresholding and normalizing each component map, followed by using the normalized cluster centroids obtained from gradient #2 k‐means clustering as the weight for the component maps to create spatial maps.

FIGURE 9

(top) Dynamic gradient ordering vectors for a healthy subject, (middle) dynamic gradient ordering associated with component #53, and (bottom) the associated inter‐component ordering synchrony plot for component #53.

FIGURE 10

The group difference map associated with inter‐component ordering synchrony plot defined as $-{\log }_{10}\left(\text{pvalue}\right)\times \mathrm{sign}\left(\text{tstatistics}\right)$ for (a) gradient #1 and (b) gradient #2. After demeaning and smoothing the index order to create inter‐component ordering synchrony plot associated with each component for each subject, the cross‐correlation across all lags is computed, followed by taking the maximum lag for each subject and comparing between patients and healthy controls. The DMN/CC/SM showed significant higher value in healthy controls in comparison with patients, however, the cross‐correlation between end components (SC/CB) were significantly lower in schizophrenia patients.

FIGURE 11

The cross‐correlation between cluster centroids for (a) HC, (b) patients with SZ, and (c) HC–SZ. The HC–SZ map is representative of the positive connectivity between the second centroid with SM and CB in controls and negative connectivity in patients for both the third centroid (SM) and the fourth centroid (CB).

FIGURE A1

The 2D spatial maps associated with each state based on the gradient #1. A spatial map of the dFNG was created by thresholding and normalizing each component map, followed by using the normalized cluster centroids obtained from the gradient #1 as the weight for the component maps to create spatial maps.

FIGURE A2

The 2D spatial maps associated with each state based on the gradient #2. A spatial map of the dFNGs was created by thresholding and normalizing each component map, followed by using the normalized cluster centroids obtained from the gradient #1 as the weight for the component maps to create spatial maps.