Review of Dynamic Resting‐State Methods in Neuroimaging: Applications to Depression and Rumination

ABSTRACT

Large‐scale functional brain networks have most commonly been evaluated using static methods that assess patterns of activation or functional connectivity over an extended period. However, this approach does not capture time‐varying features of functional networks, such as variability in functional connectivity or transient network states that form and dissolve over time. Addressing this gap, dynamic methods for analyzing functional magnetic resonance imaging (fMRI) data estimate time‐varying properties of brain functioning. In the context of resting‐state neuroimaging, dynamic methods can reveal spontaneously occurring network configurations and temporal properties of such networks. Patterns of network functioning over time during the resting state may be indicative of individual differences in cognitive‐affective processes such as rumination, or the tendency to engage in a pattern of repetitive negative thinking. We first introduce the current landscape of dynamic methods and then review an emerging body of literature applying these methods to the study of rumination and depression to illustrate how dynamic methods may be used to study clinical and cognitive phenomena. An emerging body of research suggests that rumination is related to altered functional flexibility of brain networks that overlap with the canonical default mode network. An important future direction for dynamic fMRI analyses is to explore associations with more specific features of cognition.

This timely review delves into dynamic functional connectivity approaches that could deepen our understanding of maladaptive neurocognitive processes underlying depression. Dynamic approaches may be uniquely well‐positioned to elucidate network patterns associated with repetitive negative thought such as rumination. “Dynamic” properties of large‐scale brain networks may provide insight into cognitive processes common in psychopathology. Here, we review research into abnormal network dynamics in rumination and depression, illustrating how dynamic methods may be used to study clinical and cognitive phenomena.

1. Introduction

Research has demonstrated that coordinated functioning of distributed brain systems supports cognition and behavior. Methods for evaluating large‐scale network functioning commonly focus on “static” properties, for example, dominant patterns over an extended period. However, such methods do not capture fine‐grained temporal variation in network properties, including transient fluctuations in connectivity or co‐activation patterns. Novel methods capable of evaluating these “dynamic” functional properties provide complementary insight into how the brain supports patterns of thinking. In this review, we synthesize studies utilizing dynamic functional connectivity (FC) methods to investigate neural correlates of depression and rumination, a cognitive‐affective process considered a hallmark of depression (e.g., LeMoult and Gotlib 2019).

Dynamic FC methods may be especially suited to characterize how moment‐to‐moment interactions between brain systems give rise to psychological processes central to mental illness. Due to the growing societal burden of depression, it is important to identify neural dynamics associated with cognitive‐affective processes inherent in depressive thought. Of particular interest is repetitive negative thought (RNT), a class of perseverative cognition present across most major psychiatric disorders (Ehring and Watkins 2008; Moulds and McEvoy 2025). Characterized by recurrent thoughts about past, present, or future distressing experiences (Ehring and Watkins 2008; Moulds and McEvoy 2025), RNT is difficult to control, contributes to distress and impairment, and may exacerbate other psychiatric symptoms.

Although there is debate over whether subtypes of RNT (e.g., rumination and worry) are mechanistically distinct (e.g., Puccetti et al. 2025), depression has been most closely associated with rumination, in which individuals passively focus on the causes and consequences of their psychological distress, including past negative experiences (Nolen‐Hoeksema 2000; Nolen‐Hoeksema et al. 2008; Spasojević and Alloy 2001; Koval et al. 2012). Rumination has been linked to multiple forms of psychiatric illness but shows particularly strong links to depression even after accounting for comorbidities and other forms of RNT (Hong 2007; Nolen‐Hoeksema 2000; Nolen‐Hoeksema et al. 2008; Olatunji et al. 2013; Spasojević and Alloy 2001; Koval et al. 2012; Smolker et al. 2023; Verstraeten et al. 2011). Likely emerging from interactions between cognitive, affective, and mnemonic brain systems, the neural dynamics associated with rumination remain poorly understood. Studies have begun to evaluate neural dynamics associated with depression and rumination separately, but there has been no review of this literature, making it unclear the degree to which they show overlapping dynamic functional correlates. Critically, there is a lack of consensus regarding the role of rumination in depression. Rumination is not a formal DSM‐5 symptom of depression, and yet several depressive symptoms represent forms of ruminative thought, such as worthlessness or guilt. Rumination also may be a risk factor for depression, and individuals can meet criteria for depression in the absence of rumination. The current review informs this debate by identifying dynamic neural properties shared between rumination and depression. An improved understanding of the neural correlates of depression and rumination may inform our understanding of the mechanistic overlap between the two, while also supporting the implementation of promising rumination‐focused therapies (e.g., Topper et al. 2017). In this review, we will provide a brief overview of dynamic neuroimaging methods (see also Hutchison et al. 2013; Cohen 2018; Lurie et al. 2020), discuss applications to properties of large‐scale brain networks, and evaluate initial work in depression and rumination. Specifically, we characterize the degree to which depression and rumination show overlapping dynamic FC correlates.

2. Overview of Dynamic Methods

Static methods in fMRI aim to identify functional properties of networks that are dominant over an extended period. Static measures have revealed a remarkably consistent set of large‐scale functional brain networks, reflecting communities of brain regions that tend to show correlated activity, especially during resting state (Bressler and Menon 2010; Yeo et al. 2011; Uddin et al. 2019). While static approaches have highlighted the stability of brain networks, dynamic approaches evaluate patterns of flexibility. Dynamic methods in fMRI can be defined as methodological approaches that measure time‐varying properties of functional brain networks. ¹ Hence, dynamic methods identify patterns of coordinated activity that are not well represented by the global average (i.e., more transient patterns that are still functionally meaningful, or common but opposing patterns that cancel each other out). (See Figure 1 for a conceptual overview of fMRI methods).

FIGURE 1.

Visualizing the space of fMRI analytic methods for static and dynamic network functioning. Visualizing the space of fMRI analytic methods discussed in the present review for evaluating large‐scale networks and static or dynamic properties of functioning (other methods beyond the scope of the review are not depicted). Dynamic methods are bolded, and static methods are unbolded. Along one axis, methods vary continuously in terms of the timescale of interest of the functional property they are designed to measure. For example, resting state functional connectivity generally considers patterns of connectivity that are consistent across many volumes of fMRI data (e.g., a 6+ minute scan), while methods such as LEiDA analysis consider patterns of synchrony within a single volume. Along a second axis, methods vary (categorically) in terms of whether they primarily focus on functional connectivity or activation. ^aHidden Markov Models are listed in both the Connectivity and Activation sections since they can be used to identify states defined by both properties.

2.1. Common Dynamic Properties, Measures & Methods

In this review, we focus on two domains of dynamic network properties and methods. The first is FC variability, quantifying the extent to which discrete brain regions or networks vary their synchrony with other specified regions or networks (Figure 2). The second focuses on variability in the spatial configuration and functioning of transient network states reflecting temporally constrained patterns of connectivity and/or activation (Figure 3). After transient network states are identified, dynamic properties that can be computed include frequency and sequencing of network states. Of note, these domains of dynamic properties are related: e.g., transient network states can be conceived as the ensemble of network configurations that give rise to variable FC of canonical networks over time.

FIGURE 2.

Illustration of functional connectivity variability. Functional connectivity variability refers to any quantitative measure of how BOLD signal correlations vary over time. Low functional connectivity variability indicates that regions or networks show relatively consistent patterns of connectivity over time. High functional connectivity variability indicates that regions or networks show patterns of connectivity that fluctuate over time.

FIGURE 3.

Illustration of transient network states. Transient network states represent large‐scale patterns of brain activity that reappear across time and across subjects. These can be defined based on activation, connectivity, or their derivatives. Additional metrics can be derived from patterns of state sequencing over time.

Methods used to measure FC variability of canonical networks include sliding window analysis and dynamic conditional correlation (DCC). In sliding window analysis (Sakoğlu et al. 2010), the scan time‐series is divided into a set of epochs (“windows”) that are typically 30–90 s long and may or may not overlap. A measure of brain function is computed within each window, and variability in that functional metric is computed across windows; for example, a common metric is the standard deviation of Fisher's z‐transformed Pearson's correlation coefficients for a given pair of brain regions across sliding windows. Other metrics include graph theory measures (Medaglia 2017), and the variability of these properties can be quantified using standard deviation, variance, or sample entropy (e.g., Jia et al. 2017). (See Shakil et al. 2016 for a systematic evaluation of parameter selection in sliding window analysis.) In a complementary approach, DCC avoids the use of windows by using a combination of time‐series models to derive estimates of instantaneous connectivity (for more details, see Lindquist et al. 2014). In both approaches, FC variability can be computed at several levels of spatial extent for specified networks, including: ROI‐to‐ROI (reflecting variability between a pair of regions), seed‐to‐whole‐brain (reflecting average variability between a specific region and the rest of the brain), or whole‐brain (reflecting variability between all pairs of regions).

An overlapping set of dynamic methods can be used to identify transient network states and estimate their functioning. Sliding window analyses can be used to extract window‐based transient network states reflecting time‐varying patterns of connectivity, by clustering FC matrices into a reduced set of patterns that reappear over time and across subjects. These transient network states reflect patterns of FC that emerge across the length of a window. Other dynamic methods identify framewise transient network states, aiming to increase temporal resolution by classifying brain patterns at the level of each 3D brain volume. Leading Eigenvector Dynamics Analysis (LEiDA) is one method that identifies a set of framewise states (for more details, see Cabral et al. 2017), using the Hilbert transform to compute the instantaneous phase of all brain regions in a volume. Transient network states derived from LEiDA reflect recurring patterns of whole‐brain instantaneous phase synchrony. A second approach to identifying framewise states is Co‐activation pattern (CAP) analysis (for more details, see Liu et al. 2013; Liu et al. 2018; Chen et al. 2015). CAP defines states from whole‐brain patterns of activation for each volume, yielding transient network states reflecting recurring patterns of whole‐brain co‐activation. Hidden Markov Models (HMM) can also be used to identify latent framewise states defined by activation and/or connectivity patterns (for more details, see Vidaurre et al. 2018). Across these methods for deriving transient network states, dynamic properties of interest include time‐in‐state (overall time spent in a particular state), persistence (typical duration of a given state), and transition frequencies (tendency to transition from one state to another).

2.2. Methodological Challenges and Considerations

An important debate regarding dynamic methods is whether they effectively capture signal versus noise. For example, brain networks derived from static resting state FC have been shown to be relatively consistent across people and time. Do fluctuations in these patterns of connectivity reflect functionally meaningful changes, or do they primarily reflect noise? An important line of investigation has been to evaluate the reliability of these measures over time. Several studies have found that while some dynamic measures are reliable over time, there is considerable variability, and dynamic measures tend to show lower reliability than static measures (Choe et al. 2017; Abrol et al. 2017; Zhang et al. 2018). However, it is important to note that reliability is distinct from validity; a measure with low test–retest reliability may correlate with neural or psychological processes that naturally fluctuate over time.

To evaluate the validity of dynamic measures, studies have explored associations between fMRI‐based dynamic properties and those from other modalities. Animal studies using simultaneous fMRI and electrophysiological or calcium recordings have found that BOLD FC variability correlates with changes in local field potentials and calcium‐derived measures of dynamic FC (Pan et al. 2011; Thompson et al. 2013; Thompson 2018; Matsui et al. 2019). In humans, studies using simultaneous fMRI and EEG have found associations between FC variability and changes in power across multiple frequency bands (Allen et al. 2018; Chang et al. 2013; Tagliazucchi et al. 2012). These studies provide evidence for the convergent validity of dynamic methods. Altogether, further research is required to validate distinctive dynamic measurement approaches and better delineate dynamic properties of brain function.

Each method for estimating dynamic properties of large‐scale network functioning has strengths and limitations. With the aim of measuring FC variability, sliding window analysis is widely used and is relatively straightforward to implement and interpret. However, a limitation is the arbitrary nature of the window length. Windows must be short enough to detect fluctuations in FC, but long enough to ensure that estimates are reliable. In contrast, DCC is a windowless approach, avoiding this issue, and one study found that DCC yielded dynamic metrics that showed higher test–retest reliability compared to sliding window approaches (Choe et al. 2017). Sliding window analysis is useful if researchers are interested in network states that persist on the scale of 30–90 s, but framewise approaches such as LEiDA, HMM, or CAP analysis may be more appropriate to identify states that persist for shorter durations. Finally, a challenge for most state‐based approaches is the selection of optimal clustering methods for extracting valid and reliable sets of transient network states. A difficulty with many clustering algorithms is how to select the appropriate number of clusters (Ikotun et al. 2023). Other parameters, such as distance metrics and centroid initialization in k‐means, can also influence results and have not yet been optimized (Ikotun et al. 2023). Comparing results from studies using different clustering methods can thus be difficult, especially when the number of identified states is considerably different.

Researchers interested in using dynamic methods must first identify the dynamic property of interest and then balance the strengths and limitations of methods for measuring that property. As this field grows, research directly comparing dynamic methods may further illuminate strengths and weaknesses of specific approaches.

2.3. Application of Dynamic Methods to the Study of Cognition

Dynamic properties of network functioning may confer specific insight into dynamic properties of cognitive‐affective processes central to depression. Dynamic methods are sensitive to changes in network functioning that may not align with experimental demands (e.g., task blocks) or that occur during resting state. Patterns of shifting among network states may provide insight into naturalistic shifting among cognitive modes, or a tendency to enter or dwell in specific cognitive modes.

Dynamic properties of brain function may be especially salient for understanding neural correlates of pathological thought, such as ruminative thoughts, which are persistent and difficult to disengage from (Nolen‐Hoeksema 1991). These temporal properties of rumination may correspond with changes in the temporal dynamics of functional networks, reflecting differences in how the brain “shifts gears” from moment to moment. Identifying abnormalities in the temporal functioning of large‐scale networks may provide unique insight into how and why rumination takes hold, thus providing critical insights into the neurocognitive underpinnings of depression.

Below, we review studies that explore associations of dynamic network functioning with current depression and individual differences in the self‐reported tendency to ruminate, in both clinical and nonclinical samples. (See also Sun et al. 2024). Of particular interest is understanding the degree to which dynamic connectivity markers of depression align with markers of rumination. An important caveat is that it is difficult to distinguish whether dynamic neural properties reflect brain states underlying active rumination, a propensity to engage in rumination, or some other related processes or traits. In the current review, we focus on studies evaluating associations between dynamic FC at rest and the self‐reported tendency to ruminate (Nolen‐Hoeksema and Morrow 1991). Dynamic neural markers of the self‐reported tendency to ruminate may not capture the brain states driving active rumination, but individuals who report a greater tendency to ruminate may be the most likely to actively ruminate during the resting state (Rosenbaum et al. 2017). Therefore, resting state markers of the tendency to ruminate likely capture some aspects of active rumination.

3. Depression‐ and Rumination‐Related Differences in FC Variability

Prior research has investigated (1) depression‐related differences in FC variability in currently depressed samples, and rumination‐related differences in FC variability in (2) currently depressed or (3) healthy samples (Table 1). To our knowledge, depression‐ and rumination‐related differences in FC variability have not yet been investigated in community samples.

TABLE 1.

Papers assessing depression‐ and rumination‐related differences in functional connectivity variability.

Study	Sample	Method	Spatial resolution	MDD>HC group differences	Other results
Demirtaş et al. (2016)	27 MDD, 27 HC	Hilbert transform	ROI to ROI	↓ Global FC variability ↓ FC variability of right inferior and medial frontal gyrus, left superior frontal gyrus, right posterior cingulate gyrus, and right inferior parietal lobule	—
Gao et al. (2023)	Sample 1: 100 HC Sample 2: 95 HC	Sliding window; Hurst exponent	Network to whole brain	—	Rumination: negatively correlated with FC variability among DMN regions; FC variability of the VMPFC had highest predictive value in model
Kaiser et al. (2016)	100 MDD, 109 HC	Sliding window	Seed (MPFC) to whole brain	↓ FC variability between MPFC and parahippocampal gyrus ↑ FC variability between MPFC and right insula	Rumination: correlated with FC variability between MPFC and right insula
Kaiser et al. (2018)	31 MDD, 22 HC (all female)	Sliding window	Seed (MPFC) to whole brain	—	Rumination: marginally correlated with FC variability between MPFC and left anterior insula in MDD group
Kim et al. (2023)	Training Sample: 84 HC Validation Sample: 61 HC Testing Sample: 48 HC Supplementary Sample: 61 HC Clinical Sample: 35 MDD	Dynamic conditional correlations	Seed (20 DMN regions) to whole brain	—	Rumination: associated with FC variability between DMPFC and a distributed set of regions (38 positive weights and 46 negative weights)
Long et al. (2020)	460 MDD, 473 HC	Sliding window	ROI to ROI	↑ Global FC variability ↑ FC variability of precentral gyrus, postcentral gyrus, thalamus, precuneus, posterior cingulate gyrus, and middle temporal gyrus	Depression Scores: HAMD scores correlated with global FC variability
Marchitelli et al. (2022)	35 MDD, 53 HC	Sliding window	Network to whole brain	↓ FC variability of lateral prefrontal cortex and middle frontal gyrus with all networks ↓ FC variability of ventromedial prefrontal cortex with DMN, limbic network, and salience network ↓ FC variability of insula with limbic network and dorsal attention network	Depression Scores: MADRS scores negatively correlated with FC variability of limbic network and DMN
Pang et al. (2018)	30 HC, 30 MDD, 30 BD	Sliding window	Seed (rAI) to whole brain	↓ FC variability between rvAI and right ventrolateral prefrontal cortex ↑ FC variability between rvAI and right precuneus, middle temporal pole, and left dorsolateral prefrontal cortex	—
Qiao et al. (2020)	81 MDD (43 high SI, 38 low SI), 35 HC	Sliding window	Seed (habenula) to whole brain	↓ FC variability between left habenula and right precuneus ↓ FC variability between right habenula and left angular gyrus	—
Shunkai et al. (2023)	42 Melancholic MDD, 55 HC	Sliding window	Seed (hippocampus) to whole brain	↓ FC variability between the left rostral hippocampus and left anterior lobe of the cerebellum	—
Wang et al. (2020)	51 BD, 51 MDD, 52 HC	Sliding window	Network to Network	↓ FC variability between DMN and FPN	—
Wise et al. (2017)	Primary Sample: 20 MDD, 19 HC Validation Sample: 19 MDD, 19 HC	Sliding window	ROI to ROI	↑ FC variability between MPFC and PCC	Rumination: correlated with FC variability between MPFC and PCC
Zhang et al. (2022)	53 MDD, 47 HC	Hilbert transform	Network to network	↓ Global FC variability	Rumination: anti‐correlated with global FC variability
Zheng et al. (2018)	54 MDD, 57 HC	Sliding window	ROI to ROI	↓ Global FC variability ↓ FC variability of the ACC with the rest of the brain	Depression Scores: HAMD scores correlated with global FC variability (direction unclear)
Zhou et al. (2021)	19 MDD, 22 HC	Sliding window	Seed (DLPFC) to whole brain	↓ FC variability between bilateral DLPFC and precuneus	—

Abbreviations: ACC = anterior cingulate cortex, BD = individuals with bipolar disorder, DLPFC = dorsolateral prefrontal cortex, DMN = default mode network, FC = functional connectivity, FPN = frontoparietal network, HAMD = Hamilton Depression Rating Scale, HC = healthy controls, with no psychiatric illnesses, MADRS = Montgomery‐Asburg Depression Rating Scale, MDD = individuals with major depressive disorder, MPFC = medial prefrontal cortex, PCC = posterior cingulate cortex, rAI = right anterior insula, ROI = region of interest, rvAI = right ventral anterior insula, VMPFC = ventromedial prefrontal cortex.

3.1. Depression‐Related Differences in FC Variability in Currently Depressed Samples

Research in clinical samples has shown depression‐related differences in resting state FC variability among regions of lateral prefrontal cortex (LPFC), insula, and canonical nodes of the default mode network (DMN) including medial prefrontal cortex (MPFC), posterior cingulate (PCC), and angular gyrus (Demirtaş et al. 2016; Kaiser et al. 2016; Zheng et al. 2018; Pang et al. 2018; Wang et al. 2020; Qiao et al. 2020; Long et al. 2020; Zhou et al. 2021; Shunkai et al. 2023; Marchitelli et al. 2022). The DMN is classically thought to support self‐referential processing (e.g., Hamilton et al. 2015), and research has consistently revealed depression‐related differences in FC variability of cognitive control regions including LPFC and anterior cingulate cortex (ACC) (Marchitelli et al. 2022; Zheng et al. 2018; Zhou et al. 2021), in connection with nodes of the DMN, such as MPFC (Demirtaş et al. 2016; Wang et al. 2020). These studies vary in the regions implicated and the directionality of results (i.e., whether depression is related to increased or decreased FC variability), which could be in part due to methodological differences between studies, such as ROI selection, sliding window parameters, and choice of FC variability metrics. Overall, this research suggests that disrupted flexibility in the coordination of regions involved in cognitive control and self‐directed thinking may characterize people with depression.

3.2. Rumination‐Related Differences in FC Variability in Currently Depressed Samples

A smaller number of studies have directly examined associations between FC variability and rumination in currently depressed individuals, yielding consistent evidence for altered variability in FC of brain networks involving the MPFC. Taking a sliding window approach, Kaiser et al. (2016) found that the past 2 weeks' self‐reported rumination in MDD patients was associated with higher FC variability between the MPFC and insula, a pattern driven by more frequent windows of highly correlated activity between these regions. Kaiser et al. (2018) observed similar results in an independent sample of women with MDD, finding higher variability in MPFC‐insula FC was related to negative self‐referential attentional bias and statistically mediated the positive association between rumination and negative attentional bias. Other groups have detected rumination‐related differences in the variability of MPFC FC with other regions of the canonical DMN (Wise et al. 2017), but broader blunting of whole‐brain FC variability (Zhang et al. 2022; although this analysis tested correlates of rumination after collapsing across MDD patients and healthy controls). Together, prior research shows that MPFC FC variability is associated with rumination in currently depressed individuals.

3.3. Rumination‐Related Differences in FC Variability in Healthy Samples

Consistent with studies of clinically depressed samples, research in healthy samples has identified rumination‐related differences in variability of MPFC FC with other brain regions. One investigation showed that self‐reported trait rumination in healthy subjects was best predicted by whole‐brain FC variability of DMPFC using DCC (Kim et al. 2023). They found that self‐reported rumination was related to higher FC variability between DMPFC and frontoparietal and temporal regions but decreased FC variability with other regions including occipital and cerebellar areas. In this study, models using static FC measures failed to predict rumination. Another study of healthy individuals observed rumination‐related differences in dynamic functioning of medial regions using a combination of window‐based and windowless methods (Gao et al. 2023), including lower FC variability between VMPFC and the rest of the DMN. Across studies, these results consistently show that variability in FC of medial prefrontal systems is associated with trait rumination in nonclinical samples.

3.4. Summary of Depression‐ Or Rumination‐ Related Differences in FC Variability

Research to date indicates that depression and the tendency to ruminate are associated with abnormal FC variability of the DMN. Specifically, the tendency to ruminate is associated with higher FC variability of MPFC with insula or lateral prefrontal systems, and more stability between MPFC and other DMN regions. Altogether, findings suggest that depression and rumination may both emerge from interactions involving the DMN, LPFC, and insula, brain systems supporting self‐directed thought, cognitive control, and interoception, respectively.

4. Transient Network States Related to Rumination and Depression

Prior research using state‐based approaches that identify recurring transient networks and their functioning over time has tested: depression‐related differences in (1) window‐based or (2) framewise states in current depression; (3) depression‐related differences in framewise states in community samples; and (4) rumination‐related differences in framewise states in currently depressed samples (Table 2). Current gaps in this literature include a lack of studies examining rumination‐related differences in healthy and community samples.

TABLE 2.

Papers assessing depression‐ and rumination‐ related differences in transient network states.

Study	Sample	Method	Number of states	MDD versus HC group differences	Other results
Alonso Martínez et al. (2020)	71 CMs (had experienced a breakup in the past 6 months)	LEiDA	9 (also explored 4 to 14 cluster solutions)	—	Depression Scores: The high depression group (MDI > 20) exhibited higher time‐in‐state and persistence of a prototypical DMN state. The high depression group transitioned more from the DMN state into a DMN+ state
Belleau et al. (2022)	22 HC, 14 CSA, 18 MDD, 17 MDD + CSA	CAP	8	↑ Time‐in‐state of a DMN+ state (involving co‐activation of DLPFC, PPC, angular gyrus, cerebellum, and PCC) ↑ bidirectional transitions between this state and a prototypical DMN state	Rumination: Rumination correlated with time‐in‐state of the DMN+ state. Rumination correlated with bidirectional transitions between the DMN+ state and the prototypical DMN state
Goodman et al. (2021)	445 CMs	CAP	5	—	Depression Scores: BDI scores correlated with time‐in‐state of a prototypical DMN state, for individuals with BDI ≥ 14. BDI scores negatively correlated with time‐in‐state and persistence of a DMN+ state (involving co‐activation of DLPFC, MPFC, PCC, temporal cortex, and inferior parietal lobule), for individuals with BDI ≥ 14.
Kaiser et al. (2019)	23 MDD, 22 HC	CAP	9	—	Depression Scores: CESD scores correlated with time‐in‐state and persistence of a DMN+ state (involving co‐activation of anterior insula, lateral and medial prefrontal cortex, posterior cingulate cortex, and angular gyrus). CESD scores correlated with transitions from the DMN+ state to a more prototypical DMN state Rumination: Rumination correlated with time‐in‐state and persistence of the DMN+ state. Rumination correlated with transitions from the DMN+ state to a more constrained DMN state
Karapanagiotidis et al. (2020)	256 CMs	HMM	7	—	Depression Scores: Persistence of a DMN+ state (showing co‐activation of canonical DMN and limbic regions) was associated with a factor related to rumination and depression. Rumination: Persistence of the DMN+ state was associated with a factor related to rumination and depression.
Liu et al. (2023)	50 MDD, 44 remitted MDD, 64 HC	CAP	4	↑ Time‐in‐state and entries into a DMN state (involving MPFC, PCC, angular gyrus, and middle temporal gyrus)	Rumination: Rumination correlated with number of entries into a prototypical DMN state, and time‐in‐state of an FPN‐like state
Piguet et al. (2021)	9 MDD, 22 BD, 32 HC	iCAP	19	↑ Persistence of two DMN‐like states, an insula‐centered state, and an amygdala‐centered state ↑ Number of occurrences of an insula‐centered state, and an SMN‐like state	Depression Scores: positively correlated with DMN persistence in MDD, but negatively correlated in HC Rumination: positively correlated with number of occurrences of an insula‐centered state and a posterior DMN state
Wang et al. (2020)	51 BD, 51 MDD, 52 HC	Sliding window	4	↑ Time‐in‐state and persistence of a state involving relatively high rCEN‐SN connectivity, moderate SN‐pDMN connectivity, and strong anti‐correlations between the aDMN with SN, rCEN, and pDMN	—
Wu et al. (2019)	109 MDD, 107 HC	Sliding window	4	↓ Time‐in‐state and persistence for a state involving strong connectivity within and between visual network, auditory network, and somatomotor network regions, along with moderate connectivity between these networks and DMN	Depression Scores: BDI scores negatively correlated with time‐in‐state and persistence of a state involving positive connectivity within and between visual network, auditory network, and somatomotor network regions
Yang et al. (2022)	48 MDD, 46 HC	Sliding window	6	↑ Time‐in‐state of states involving relatively weak and uniform connectivity across the whole brain ↑ Time‐in‐state of a state involving positive connectivity within and between visual network, somatomotor network, and cerebellum, and negative connectivity between FPN and those regions ↓ Time‐in‐state of a state involving mostly positive connectivity, particularly between and within DMN, visual network, and somatomotor network regions	Depression Scores: HAMD scores negatively correlated with time‐in‐state and persistence of a weakly connected state
Zhi et al. (2018)	182 MDD, 218 HC	Sliding window	5	↑ Time‐in‐state of a state involving weak connectivity overall ↓ Time‐in‐state of a state involving positive connectivity within visual network and between visual network and DMN ↓ Time‐in‐state of a state involving positive connectivity within and between visual network and somatomotor network regions	—

Abbreviations: aDMN = anterior default mode network, BD = individuals with bipolar disorder, BDI = Beck Depression Inventory‐II, CAP = co‐activation patterns, CESD = Center for Epidemiological Studies‐Depression, CM = community member, eligibility not dependent on mood disorder status, CSA = individuals with a history of childhood sexual abuse, DMN = default mode network, FPN = frontoparietal network, HAMD = Hamilton Depression Rating Scale, HC = healthy controls, with no psychiatric illnesses, HMM = Hidden Markov Models, iCAP = innovation‐driven co‐activation patterns, LEiDA = leading eigenvector dynamics analysis, MDI = Major Depression Inventory, MPFC = medial prefrontal cortex, PCC = posterior cingulate cortex, pDMN = posterior default mode network, MDD = individuals with major depressive disorder, rCEN = right central executive network, SMN = sensorimotor network, SN = salience network.

4.1. Depression‐Related Differences in Window‐Based States in Currently Depressed Samples

Several prior studies have used sliding window analyses to identify transient network states and test their associations with depression status. Results indicate that individuals with MDD spend more time than non‐depressed individuals in whole‐brain states characterized by weaker FC (Wang et al. 2020; Zhi et al. 2018; Yang et al. 2022), including more time in states involving weak positive connectivity between canonical nodes of the frontoparietal network, including LPFC and posterior parietal cortex (PPC), and DMN regions such as MPFC and PCC (Wang et al. 2020; Zhi et al. 2018; Yang et al. 2022). In a separate study (Wu et al. 2019), depressed individuals spent less time in a strong connectivity state involving positive FC among regions canonically associated with the somatomotor network (precentral and postcentral gyrus), visual network (occipital cortex), and auditory network (superior temporal gyrus). Together, these studies provide evidence that individuals with MDD spend more time in transient network states defined by weaker FC, especially of frontoparietal regions, which may in turn be driven by lower positive or more variable patterns of FC.

4.2. Depression‐Related Differences in Framewise States in Currently Depressed Samples

Methods for measuring transient network states based on single‐volume patterns of BOLD signal provide insight into coordinated activity or FC across the brain that may emerge and persist over short or long timescales. Several studies applying CAP analysis to resting state have observed multiple transient network states involving DMN regions (Kaiser et al. 2019; Belleau et al. 2022; Piguet et al. 2021; Janes et al. 2019). Results consistently identify a state mostly limited to canonical DMN regions such as MPFC, PCC, and angular gyrus, but also one or more hybrid DMN+ states involving co‐activation of canonical DMN regions with regions not typically considered part of the DMN, especially LPFC and anterior insula (Kaiser et al. 2019; Belleau et al. 2022; Piguet et al. 2021; Janes et al. 2019). These studies have found that currently depressed individuals transition more frequently between a canonical DMN state and DMN+ states, and spend more time overall and persist for longer in DMN+ states (Kaiser et al. 2019; Belleau et al. 2022; Piguet et al. 2021). One study showed that currently depressed individuals spent more time in a canonical DMN state (Liu et al. 2023), although that study used a coarser clustering solution yielding only one DMN‐like state, and hence hybrid DMN+ states may not have been detected. Overall, results indicate that depressed individuals spend more time in states involving co‐activation of canonical or hybrid DMN+ states.

4.3. Depression‐Related Differences in Framewise States in Community Samples

Studies examining transient network states in nonclinical samples identified DMN+ states similar to those observed in clinical samples, including frontoinsular DMN+ and frontolimbic DMN+ states (Alonso Martínez et al. 2020; Goodman et al. 2021; Karapanagiotidis et al. 2020). Consistent with findings in clinical samples, higher severity of depressive symptoms in nonclinical samples was associated with more frequent entries into frontoinsular DMN+ states (but lower total time in these states) (Alonso Martínez et al. 2020; Goodman et al. 2021) and increased persistence of frontolimbic DMN+ states (Karapanagiotidis et al. 2020). Depression‐related increases in dominance of canonical DMN states were also observed in nonclinical samples, for example, increased time spent in states of co‐activation or connectivity of MPFC, PCC, inferior parietal lobule, and middle temporal gyrus (Alonso Martínez et al. 2020; Goodman et al. 2021). Together, these results in nonclinical samples indicate consistent associations between depressive symptoms and altered overall time in, persistence of, and transitions among states involving DMN regions. Notably, depression‐related differences in transient network functioning are similar across clinical and community samples.

4.4. Rumination‐Related Differences in Transient Framewise States in Currently Depressed Samples

Increased time spent in frontoinsular DMN+ states involving co‐activation of canonical DMN regions extending to LPFC and/or anterior insula has been consistently related to higher levels of self‐reported rumination in samples involving currently depressed individuals (Kaiser et al. 2019; Belleau et al. 2022; Piguet et al. 2021). Several studies have also linked the tendency to ruminate with increased transitions between hybrid DMN+ states and canonical DMN states (Kaiser et al. 2019; Belleau et al. 2022). It is notable that the number of transient network states—and hence the number of potential rumination‐related states—varies by clustering solution across studies, emphasizing the importance of attending to methodological differences in network dynamics research. Piguet et al. (2021) extracted 19 states and identified two DMN+ rumination‐related states: one involved co‐activation of anterior insula, VMPFC, and ACC; the other primarily involved posterior regions of the canonical DMN (including PCC, angular gyrus, and middle/superior temporal gyrus). The conjunction of these two states bears a striking resemblance to the frontoinsular state identified by Kaiser et al. (2019), who extracted eight states. Comparing transient networks across datasets and methodological approaches is useful not only for identifying consistent patterns of rumination‐related differences, but also for informing best practices in dynamic network methods.

4.5. Summary of Depression‐ Or Rumination‐Related Differences in Transient Network States

Depression and rumination are consistently associated with increased transitions to and time in transient network states involving prefrontal and canonical DMN (midline and parietal) regions. Both depression diagnoses and symptom levels in community samples were associated with increased dominance of DMN/DMN+ networks and transitions among canonical and hybrid DMN+ states. Among depressed individuals, those reporting higher levels of rumination especially show increased dominance of frontoinsular DMN+ states. Depression was also associated with increased time spent in (window‐based) states of weaker frontoparietal FC, a pattern that may in turn be driven by increased dominance of DMN+ states that recruit frontoparietal regions in co‐activation with canonical DMN regions.

5. Discussion

Dynamic properties of large‐scale network functioning hold promise for illuminating neural mechanisms behind pathological cognitive–affective processes in depression, with potential for future clinical application. Static imaging approaches have well‐established that increased activity and connectivity of DMN regions correlate with depression and rumination (e.g., Hamilton et al. 2015; Zhou et al. 2020), and the emerging dynamic literature builds on and complements these findings with consistent evidence that depression and the tendency to ruminate are both associated with disrupted FC variability of regions canonically associated with the DMN (Demirtaş et al. 2016; Pang et al. 2018; Wang et al. 2020; Qiao et al. 2020; Long et al. 2020; Zhou et al. 2021; Shunkai et al. 2023; Marchitelli et al. 2022). Specifically, multiple dynamic methods suggest that depression and self‐reported rumination are both linked to alterations in the neural dynamics between the MPFC and other DMN areas, as well as the LPFC and the insula (see Figure 4) (Demirtaş et al. 2016; Kaiser et al. 2016, 2018; Wise et al. 2017; Wang et al. 2020; Gao et al. 2023; Kim et al. 2023). Across studies, there is considerable agreement that depression and rumination are both associated with higher variability in dynamic connectivity of the DMN with other regions, as well as increased time spent overall and persistence of transient DMN+ networks involving areas of canonical DMN, insula, and (sometimes) LPFC or limbic systems (Kaiser et al. 2019; Belleau et al. 2022; Piguet et al. 2021; Karapanagiotidis et al. 2020). Notably, this pattern has been remarkably consistent even across different methods for defining transient network states (i.e., LEiDA and CAP analysis) and across different numbers of transient network states. Together these results highlight changes in dynamic properties of the DMN, frontoparietal network, and the insula as corresponding with both depression diagnosis, symptom levels, and the tendency to ruminate in both depressed and non‐depressed individuals, suggesting that the interactions between the brain systems may drive the close relationship between depression and rumination.

FIGURE 4.

Preliminary brain regions and networks associated with altered fMRI dynamics in depression and rumination. Illustration of brain regions and associated canonical networks that frequently show differences in dynamic resting state fMRI metrics based on depression status or self‐reported rumination.

How might we interpret evidence that rumination is consistently related to abnormalities in the dynamic functioning of networks involving canonical DMN systems? The DMN has been associated with introspection and internally directed cognitive processes (Andrews‐Hanna et al. 2013), while the insula and LPFC are thought to help regulate the activity of other large‐scale networks (Menon and Uddin 2010; Miller and Cohen 2001). Dominance of frontoinsular‐DMN networks and variability in FC among these systems may reflect a top‐down bias towards allocating resources to maladaptive introspection, ineffective efforts to regulate introspection, and/or bottom‐up hijacking of network‐control systems by regions involved in introspection. Additional research is needed to explicitly test these hypotheses.

At this early stage of development in dynamic network research, several areas of methodological uncertainty add challenge to interpreting results. The relationship between dynamic properties is not well understood, and it is challenging to understand the extent to which dynamic properties reflect “state”‐ or “trait”‐like processes. In the current review, we focused on associations with self‐reported tendencies to ruminate, which have been described as more trait than state‐like. However, the neural correlates of the tendency to ruminate need not be the same as the brain states underlying actual rumination, and it is important for future research to apply dynamic connectivity methods to rumination induction paradigms, in which participants actively engage in rumination. To the extent that dynamic properties observed at rest reflect state‐like features of cognitive phenomena, they can aid in our ability to better categorize the various cognitive processes occurring during resting state scans, or to predict trial level outcomes (see Gonzalez‐Castillo and Bandettini 2018 for review). Dynamic methods may also capture trait differences in large‐scale network functioning, such as how a person tends to move through network states. Future research that evaluates the regularity of network dynamics across various timescales or task paradigms can help us understand how these properties change over time or with external demands.

Additional challenges limiting interpretation of the reviewed literature are specifically related to the study of trait rumination and depression. First, depression is heterogeneous, and specific depressive phenotypes could be related to both distinct patterns of network dynamics and differing levels of rumination. Second, changes in network dynamics may diverge between clinical and subclinical populations. For example, subclinical depression was associated with dominance of canonical DMN network states (Alonso Martínez et al. 2020; Goodman et al. 2021), but clinical depression was associated with a hybridized DMN+ state also involving LPFC and/or insula (Kaiser et al. 2019; Belleau et al. 2022; Piguet et al. 2021). One possibility is that the more classic DMN state may dominate in earlier stages or milder presentations of depression, and this state could act as a gateway to the frontoinsular DMN state that tends to characterize more severely depressed samples. Limited research has explored how network states change longitudinally over the course of depression, which could provide further insights (see Kaiser et al. 2022; Figueroa et al. 2019; Liu et al. 2023). Despite challenges, there is emerging evidence that these dynamic approaches may have potential for intervention, particularly in the context of predictive modeling (e.g., Pilmeyer et al. 2024) or neurofeedback (e.g., Ganesan et al. 2025).

In conclusion, the study of dynamic properties of large‐scale network functioning is a growing area with the potential to provide promising insight into cognition and psychopathology. The present review focused on depression and the self‐reported tendency to ruminate, but in future research, dynamic fMRI analyses may help explain other processes important for psychiatric health such as mood, attention, arousal, and mind wandering (Betzel et al. 2017; Nummenmaa et al. 2014; Tobia et al. 2017; Cavanna et al. 2018; Mooneyham et al. 2017; Mittner et al. 2014). Dynamic methods complement traditional static methods, with the potential to provide new information about how large‐scale network functioning relates to cognition, cognitive style, and psychopathology.

Author Contributions

E.C.P. drafted the paper, and H.R.S., A.D.M., and R.H.K. provided critical revisions.

Disclosure

The authors have nothing to report.

Ethics Statement

The authors have nothing to report.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Peterson, E. C. , Smolker H. R., Moser A. D., and Kaiser R. H.. 2025. “Review of Dynamic Resting‐State Methods in Neuroimaging: Applications to Depression and Rumination.” Human Brain Mapping 46, no. 15: e70377. 10.1002/hbm.70377.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analysed during the current study.