Table 1.
Summary of studies assessing time-varying resting state functional connectivity in healthy subjects and simulated data.
| Study | RS fMRI acquisition parametersΩ | TVC analysis approach∧ | Study subjectsθ | Main findings |
|---|---|---|---|---|
| Allen et al. (2014) | Siemens Trio 3T 152 volumes TR = 2 s |
1. Group ICA decomposition in 50 relevant independent components of interest, classified into 7 different functional networks 2. Sliding-window analysis, window length = 22 TRs (44 s), steps = 1 TR (2 s). 3. k-means clustering (7 recurring states) |
405 healthy adults 200 females (49.4%) mean age = 21.0 years age range = 12–35 years |
- Identification of recurring TVC states that partially diverge from static connectivity patterns - Regions belonging to the DMN have highly variable connectivity over time, while regions of the sensory and motor networks exhibit more stable connectivity configurations |
| Allen et al. (2017) | Siemens Sonata 1.5T Eyes open/Eyes closed 255 volumes TR = 2 s |
1. Group ICA decomposition in 43 relevant independent components of interest, classified into seven different functional networks 2. Sliding-window analysis, window length = 30 TRs (60 s), steps = 1 TR (2 s) 3. k-means clustering (5 recurring states) 4. Correlations with EEG data |
23 healthy adults 7 females (30.4%) mean age = 29 years SD = 8.8 years |
- States were replicable with those of Allen et al. (2014) - TVC states correspond to neurophysiological mental states detected with EEG - Eyes open/eyes closed conditions show some common and some different connectivity patterns Connectivity between the thalamus and the cortex changes from positive to negative in eyes closed vs. open condition |
| Cabral et al. (2017) | Siemens Avanto 1.5T 180 volumes TR = 2 s |
1. Segmentation in 90 cortical brain regions of the AAL atlas 2. Phase-coherence connectivity at each time point 3. Leading eigenvectors and subsequent k-means clustering (five recurring states) |
55 healthy adults with good cognitive performance 31 females (44.6%) mean age = 64 years SD = 9 years 43 healthy adults with poor cognitive performance 29 females (66%) mean age = 66 SD = 8 years |
- More frequent switches in subjects with poor cognitive vs. good cognitive performances - The lower occurrence of a state of global, positive coherence is associated with worse cognitive performances |
| Cai B. et al. (2018) | Siemens Trio 3T 126 volumes TR = 3 s |
1. Segmentation in 264 regions of the Power atlas (Power et al., 2011), grouped into 10 functional networks 2. Sliding-window analysis, window length = 50 TRs (150 s), steps = 1 TR (3 s) and dynamic sparse connectivity models 3. k-means clustering analysis (4 recurring states) |
Philadelphia neurodevelopmental cohort database 240 young adults 146 females (60.8%) mean age = 18.99 years SD = 1.12 years 232 children 123 females (53%) mean age = 10.67 years SD = 1.09 years |
- Compared with young adults, children had increased connectivity between the DMN and other subnetworks - Children had reduced connectivity among sensorimotor, executive control and auditory networks vs. young adults - Young adults spent more time in the most connected state |
| Chang and Glover (2010) | GE Signa HDx or Signa 750 3T 360 volumes TR = 2 s |
1. ROIs in crucial nodes of the DMN and of the “task-positive” (executive control) network 2. Time-frequency decomposition using Wavelet transform coherence; sliding-window analysis |
12 healthy adults 6 females (50%) mean age = 27.7 years SD = 12.4 years |
- Coherence and phase between the PCC and nodes of the executive control network significantly vary in time and frequency - High variability over time was observed between the PCC and brain areas involved in higher-level cognitive functions |
| Chen T. et al. (2016) | Siemens Skyra 3T Eyes open 1,200 volumes TR = 0.72 s Test-retest data |
1. Segmentation in 264 regions of the Power atlas (Power et al., 2011) 2. Sliding-window analysis, window length = 55 TRs (40 s), steps = 1 TR (0.72 s) 3. Graph theoretical analysis |
Human Connectome Project dataset 77 healthy adults 50 females (64.1%) age range = 22–35 years |
- The salience network showed highly flexible connectivity with fronto-parietal, cingulate-opercular, and attention networks - The salience network maintained a consistently high level of network centrality over time |
| Choe et al. (2017) |
Multi-Modal MRI Reproducibility Resource (Kirby) data set
Philips Achieva 3T 210 volumes TR = 2 s Test-retest data Human Connectome Project S500 Data dataset Siemens Skyra 3T 1,200 volumes TR = 0.72 s Test-retest data |
Kirby dataset: 1. Group ICA decomposition in 39 relevant components of interest, classified into 7 functional networks 2. Sliding-window analysis, window length = 30 TRs (60 s) Human Connectome Project S500 Data dataset: 1. Group ICA decomposition in 50 relevant components of interest 2. Sliding-window analysis, window lengths = 15, 30, 60, and 120 TRs (11, 22, 43, and 86 s) 3. TVC mean and variance, k-means clustering (three recurring states) and dynamic conditional correlation approaches |
Kirby dataset 20 healthy adults Human Connectome Project S500 Data dataset 523 healthy adults |
- TVC can be reliably estimated in test-retest data - The dynamic conditional correlation method seems to be more reliable than sliding-window analysis |
| Lim et al. (2018) | Siemens Prisma 3T Eyes open 250 volumes TR = 2 s |
1. Segmentation of 114 regions of the Yeo atlas (Yeo et al., 2011), classified into 17 functional networks 2. Sliding-window analysis, window length = 7 TRs (14 s), steps = 1 TR (2 s) 3. k-means clustering (3–7 recurring states) |
21 healthy adults with high-trait mindfulness 13 females (61.9%) mean age = 23.7 years SD = 3.4 years 18 healthy adults with low-trait mindfulness 13 females (72.2%) mean age = 21.9 years SD = 2.3 years |
- High trait mindfulness subjects spent significantly more time in a high within-network connectivity state, characterized by greater anti-correlations between task-positive networks and the DMN - Transitions between brain states was more frequent in high vs. low trait mindfulness subjects |
| Lindquist et al. (2014) | Philips Achieva 3T 210 volumes TR = 2 s Test-retest data |
1. Segmentation of six spherical ROIs (radius = 3 mm) containing regions of the DMN 2. Point-process analysis 3. Estimation of variance of dynamic connectivity correlations, compared with traditional sliding-window analysis |
Multimodal MRI Reproducibility Resource (Kirby21) dataset 21 healthy adults 10 females (47.6%) mean age = 31.76 years SD = 9.47 years |
- Dynamic conditional correlations are able to quantify dynamics of RS fMRI data - Dynamic conditional correlations have a similar performance as sliding-window analysis in quantifying TVC between brain regions |
| Liu and Duyn (2013) | Multicenter 3T scanners Volumes varying from 119 to 195 TR varying from 2.3 to 3 s |
1. Segmentation of two spherical ROIs (radius = 6 mm) containing the PCC and left intraparietal sulcus 2. Point-process analysis 3. k-means clustering of coactivation patterns (eight coactivation patterns for the PCC and 12 for the left intraparietal sulcus) |
1000 Functional Connectomes Project (FCP) 247 healthy adults 151 females (61.1%) mean age = 22.72 years SD = 4.61 years age range = 18–44 years |
- Point-process analysis was able to extract correlational patterns in RS fMRI data from relatively brief periods of co-activation (or co-deactivation) of brain regions - Co-activation patterns resembled classical networks derived from static RS FC analysis, while more fine-grained co-activation patterns were detected |
| Marusak et al. (2017) | GE Signa 3T Siemens Verio 3T both scanners: 180 volumes TR = 2 s |
1. Group ICA decomposition in 25 relevant independent components of interest, classified into 3 functional networks 2. Sliding-window analysis, window length = 22 TRs (44 s), steps = 1 TR (2 s) 3. k-means clustering (six recurring states) 4. Correlation with age and internal thoughts |
Stanford University dataset 73 normally developing children 34 females (46.57%) mean age = 12.47 SD = 1.88 years Wayne State University dataset 73 normally developing children 49 females (67.12%) mean age = 12.09 years SD = 2.54 years |
- The occurrence and amount of time spent in specific TVC states are related to the content of self-generated thought during the scan - Temporal variability of TVC among cognitive networks increases with age - Regions showing the highest TVC include multi-modal areas associated with high-order cognitive functions, such as the precuneus and inferior parietal lobe |
| Marusak et al. (2018) | Siemens Verio 3T 390 volumes TR = 1.5 s |
1. Group ICA decomposition in four relevant independent components of interest 2. Sliding-window analysis, window length = 30 TRs (45 s), steps = 1 TR (1.5 s) 3. k-means clustering (5 recurring states) 4. Correlations with mindfulness scores |
42 children 23 females (54.8%) mean age = 10.3 years SD = 2.9 years age range = 6–17 years |
- High-mindfulness children had a greater number of transitions between states than low-mindfulness children and showed a state-specific reduction in connectivity between salience/emotion and central executive networks |
| Nini et al. (2017) | Siemens Trio 3T 225 volumes TR = 2.48 s |
1. Segmentation in 90 regions of the AAL atlas 2. Sliding-window analysis, window length = 25 s, steps = 0.6 s 3. Graph theory analysis: flexibility and variance |
1,000 Functional Connectomes Project 148 healthy young adults 74 females (50%) age range = 18–26 years |
- Flexibility of amygdala, hippocampus, fusiform gyrus, and temporal gyrus was higher in males than in females - Flexibility of middle cingulate cortex, thalamus, precuneus, and temporo-occipital regions was higher in females than in males |
| Shi et al. (2018) | Siemens Trio 3T 232 volumes TR = 2 s |
1. Group ICA decomposition in 5 relevant independent components of interest 2. Sliding-window analysis, window length = 30 TRs (60 s), steps = 1 TR (2 s) 3. k-means clustering (four recurring states) and fuzzy-meta states analyses |
Southwest University Longitudinal Imaging Multimodal dataset 331 healthy young adults 247 females (74.6%) mean age = 20.20 years SD = 1.34 years 212 healthy young adults 115 females (54.2%) mean age = 22.36 years SD = 1.49 years |
- Subjects having a high score in subjective well being spent less time in a state characterized by low cross-network connectivity and strong within-network connectivity - The total number of transitions across states was correlated with a higher subjective well-being score |
| Smith et al. (2018) | Siemens Skyra 3T scanner Eyes open 1,200 volumes TR = 0.72 s Test-retest data |
1. Segmentation of 90 regions from Shirer et al. (Shirer et al., 2012) 2. Point-process analysis 3. k-means clustering of coactivation patterns (four recurring states) |
Human Connectome Project S500 Data dataset 100 healthy adults 54 females (54%) |
- Brain state- properties were reliable across days - Summary metrics of brain connectivity dynamics had an adequate test-retest reliability |
| Tagliazucchi et al. (2013) | Siemens Trio 3T 1,505 volumes TR = 2.08 s |
1. Group ICA decomposition in six relevant independent components of interest 2. Detrended fluctuation analysis 3. Hurst exponent (measuring long-range temporal dependence) |
39 healthy adults | - Temporal memory of RS fMRI time series decreases from wakefulness to deep non-rapid eye movement sleep - Long-range temporal dependence decreases especially in regions of the DMN and attention network |
| Vidaurre et al. (2018) |
Human Connectome Project dataset
Siemens Skyra 3T Eyes open 1,200 volumes TR = 0.72 s UK Biobank dataset Siemens Skyra 3T Eyes open 490 volumes TR = 0.735 s |
1. Group ICA decomposition in 50 relevant independent components of interest from the HCP dataset, in 55 relevant independent components of interest from the UK Biobank dataset 2. Hidden Markov model 3. Stochastic inference (12 recurring states) |
Human Connectome Project dataset 820 healthy adults 453 females (55.2%) age range = 22–35 years UK Biobank dataset 5847 healthy adults age range = 40–69 years |
- Hidden Markov models allow to model resting (or task-related) brain activity as a time-varying sequence of distinct brain networks, also when analyzing very large amounts of data |
| Yaesoubi et al. (2015a) | Data from Allen et al., 2014
Siemens Trio 3T 152 volumes TR = 2 s |
1. Group ICA decomposition in 50 relevant independent components of interest Time-frequency decomposition 2. k-means clustering (five recurring states) |
Data from Allen et al. (2014) 405 healthy adults 3. 200 females (49.4%) 4. mean age = 21.0 years 5. age range = 12–35 years |
- A new time-frequency decomposition approach, based on wavelet transform coherence, detected time-frequency connectivity variations in RS fMRI data - Recurring connectivity patterns in time-frequency domain revealed significant between-group differences based on sex |
| Yaesoubi et al. (2015b) | Data from Allen et al. (2014) Siemens Trio 3T 152 volumes TR = 2 s |
1. Group decomposition in 50 relevant components of interest, classified into seven different functional networks 2. Sliding-window analysis, window length = 32 TRs (44 s), steps = 1 TR (2 s) 3. Clustering of sliding-window matrices using temporal ICA to find maximally mutually temporally independent connectivity patterns (five recurring states) 4. Sex differences |
Data from Allen et al. (2014) 405 healthy adults 200 females (49.4%) mean age = 21.0 years age range = 12–35 years |
- A method alternative to k-means clustering is proposed, based on temporal ICA. This method allowed to detect temporally independent connectivity states - Frequency of occupancy of such states was not different between genders |
| Yaesoubi et al. (2017b) | Data from Allen et al. (2014) Siemens Trio 3T 152 volumes TR = 2 s |
1. Group ICA decomposition in 50 relevant independent components of interest 2. Time-frequency decomposition 3. k-means clustering of z-scored time-frequency decompositions to find recurring frequency modes (four recurring modes) |
Data from Allen et al. (2014) 405 healthy adults 200 females (49.4%) mean age = 21.0 years age range = 12–35 years |
- Time-frequency decomposition allowed to capture frequency variations in individual network time courses - Frequency modes represent “periodic” activities consisting of instantaneous activations and deactivations |
| Yang et al. (2014) | Siemens Trio 3T 884 volumes TR = 0.645 s Test-retest data |
1. Four spherical ROIs (radius = 3 mm) in crucial nodes of the posteromedial cortex; segmentation of 156 regions from Craddock et al. (2012) 2. Sliding-window analysis, window length = 69 TRs (44 s), steps = 3 TRs (2 s) 3. Hierarchical clustering (five recurring states) |
22 healthy adults 4. 6 females (27.3%) 5. mean age = 33.5 years 6. SD = 12.5 years 7. age range = 19–60 years |
- Each subregion of the posteromedial cortex was associated with five recurring connectivity states Each subregion possessed a unique preferred state and distinct transition patterns |
| Zalesky et al. (2014) | Siemens Skyra 3T 1,200 volumes TR = 0.72 s |
1. Segmentation in different numbers of ROIs (from 90 to 4,000) (Zalesky et al., 2010) 2. Sliding-window analysis, window length = 60 s, steps = 1 TR (0.72 s) 3. Non-stationarity of RS fMRI fluctuations measured using an ad hoc test statistic |
Human connectome project Q2 Data dataset 10 healthy adults 6 females (60%) age range = 22–35 years |
- A consistent set of functional connections had pronounced fluctuations over time - The most dynamic connections were inter-modular and involved hubs of the DMN and fronto-parietal network |
| Zhang C. et al. (2018) | Siemens Skyra 3T 1,200 volumes TR = 0.72 s Test-retest data |
1. Segmentation in 116 regions of the AAL atlas and 160 regions of the Dosenbach atlas (Dosenbach et al., 2010) Sliding-window analysis, window length = from 20 TRs to 200 TRs 2. Standard deviation from the mean and excursion from the median. Amplitude of low-frequency fluctuations across sliding windows |
Human connectome project S900 Data dataset 820 healthy adults 454 females (55.4%) age range = 22–37 years |
- TVC was reliable, especially when windows size was between 30 and 50 TRs, but less reliable than static FC - The highest reliability for static and dynamic FC analysis was found for intra-network connections in the fronto-parietal, DMN, sensorimotor, and occipital networks |
Ω
All RS scans were acquired in the eyes-closed condition, except where indicated.
∧
TVC analysis approach summarizes: (1) ROIs used; (2) assessment of time-varying correlations between brain regions; (3) features extracted for assessing TVC.
θ
For each study group of healthy subjects, sex is represented as number of females (%), mean age and standard deviation (SD).
RS, resting state; fMRI, functional magnetic resonance imaging; TVC, time-varying functional connectivity; ICA, independent component analysis; TR, repetition time; SD, standard deviation; EEG, electroencephalographic registration; AAL, automated anatomical labeling; ROIs, regions of interest; DMN, default-mode network; PCC, posterior cingulate cortex; HCP, Human connectome project; UK, United Kingdom; FC, functional connectivity.