Skip to main content
NCBI home page
Psychoradiology logoLink to Psychoradiology
. 2026 Jan 5;6:kkag001. doi: 10.1093/psyrad/kkag001

Personalized functional network connectivity abnormalities in chronic insomnia disorder

Xiaoxuan Li 1, Yiran Zhai 2, Tianwei Qin 3, Zhiwei Zhang 4, Zixi Zhao 5, Hanbin Deng 6,✉, Siqi Yang 7,✉, Jiang Zhang 8,✉, Liang Gong 9,✉
PMCID: PMC12947161  PMID: 41767428

Abstract

Background

Chronic insomnia disorder (CID) is associated with disrupted functional brain networks, yet prior research has focused primarily on group-level analyses. This study employed personalized functional network mapping to identify connectivity abnormalities in CID.

Methods

Resting-state functional magentic resonance imaging (rs-fMRI) data were collected from 86 CID patients and 38 good sleeper controls (GSCs). Using non-negative matrix factorization (NMF), we derived individualized large-scale brain networks for each participant to uncover subject-specific connectivity changes in CID. We also constructed functional network connectivity (FNC) matrices using Pearson correlation coefficients and compared global and local graph-theory metrics across groups based on these individualized networks.

Results

FNC analysis revealed significant differences between CID patients and GSCs within the default mode network (DMN), ventral attention network, visual network (VIS), and other key brain regions. CID exhibited altered global network topology and significant differences in local topological properties. At the global level, CID demonstrated significantly higher small-worldness (Sigma) and normalized clustering coefficient (Gamma). At the nodal level, CID showed increased local efficiency and clustering coefficient, as well as decreased nodal efficiency in the DMN, along with increased degree centrality in the VIS.

Conclusion

By focusing on individualized functional connectivity, this approach reveals unique “fingerprint” alterations in CID. These findings provide novel insights into CID’s neurobiological mechanisms and underscore the value of personalized network approaches for understanding and treating sleep disorders.

Keywords: chronic insomnia disorder, functional network connectivity, non-negative matrix factorization, resting-state functional magnetic mesonance imaging, small world

Introduction

Insomnia is a common neuropsychiatric disorder (Mayer et al., 2021). The most common symptoms of insomnia are difficulty in falling asleep or in maintaining sleep (Tursunboyev et al., 2025). Occasional insomnia due to external stimuli is a normal phenomenon, but chronic insomnia disorder (CID) can lead to decreased sleep quality and reduced sleep duration, resulting in issues such as daytime lack of concentration and memory impairment (National Institutes of Health, 2005). CID is defined as when it occurs at least three times a week and persists for more than 3 months (Ruoff and Rye, 2016). CID not only causes persistent fatigue that undermines quality of life (Balay-Dustrude and Shenoi, 2023) and raises the risk of depression and anxiety (Hertenstein et al., 2019; Wu et al., 2024), but also drives heavy healthcare use and costs (Streatfeild et al., 2021; Chaput et al., 2023), making it vital to investigate its neurobiological roots and consequences (Spiegelhalder et al., 2013).

Previous studies have focused on exploring the neurobiological mechanisms of CID, but the understanding of brain network mechanisms remains incomplete. Functional magnetic resonance imaging (fMRI) is a noninvasive brain imaging method (Lunkova et al., 2021; Sun et al., 2024; Zhang et al., 2025). Resting-state fMRI (rs-fMRI) uncovers intrinsic functional networks such as the default mode network (DMN), salience network (SN), and executive control networks (Seeley et al., 2007). Functional network connectivity (FNC) is one of the most important analytical perspectives applying to CID. Most brain imaging analyses have been conducted at the group level, providing limited insight into the individualized neural network mechanisms of chronic insomnia. Nie et al. (2015) applied Pearson correlation to examine aberrant connectivity strengths between paired DMN subregions, demonstrating that individuals with CID show disrupted FC patterns within this network. Li et al. (2018a) employed voxel-based morphometry (VBM). Their findings illuminated heightened linkages within the left anterior cingulate cortex (ACC)/insula, left posterior cingulate, and the right limbic lobe/cingulate gyrus/paracingulate gyrus with the ACC in CID. A recent study employed a combination of independent component analysis (ICA) and a dual regression approach. The findings revealed that in individuals with CID, there were observed functional connectivity abnormalities within the SN (Cheng et al., 2022).

While group-level approaches are commonly employed for fMRI investigations, it is important to note that the average brain representation at the group level cannot fully encapsulate the intricacies of an individual’s brain (Mueller et al., 2013; Langs et al., 2016). In other words, FC can exhibit unique, fingerprint-like patterns that enable identification of individuals within a cohort (Finn et al., 2015). In recent research, Laumann et al. (2015) have indicated that functional systems exhibit a substantial degree of similarity between individuals and groups, but certain features within individuals display distinct topological characteristics. Clinically, alterations in these metrics have been associated with cognitive performance and symptom severity in neuropsychiatric disorders, and serve as potential biomarkers for early diagnosis and personalized treatment planning (Crossley et al., 2014). To accurately distinguish brain activity in CID from that in good sleeper controls (GSCs), we need to refine fMRI data processing and take individual variability into account (Wang et al., 2015). In this context, the value of individualized functional networks is 2-fold: (i) they preserve trait-like, subject-unique connectivity patterns that support reliable identification and behavior prediction (Finn et al., 2015; Cai et al., 2021), and (ii) they reduce template-misalignment bias that can blur case-control effects when functional topography varies across individuals.

To address this gap, we adopt an individualized network mapping strategy based on non-negative matrix factorization (NMF). NMF produces clear and interpretable brain maps for each subject. It decomposes data into additive, non-negative components that form sparse, parts-based representations, improving interpretability relative to unconstrained decompositions (Lee and Seung, 1999). Spatially regularized, subject-level NMF has been shown to produce individualized, coherent large-scale functional networks (Zhang et al., 2022, 2024). Importantly, individual network topography derived with these methods relates to brain maturity and executive function, highlighting their biological relevance (Cui et al., 2020). NMF offers an advantage over methods such as group-ICA with back-reconstruction, which can constrain individual maps too strongly to the group average, potentially obscuring subject-specific features (Bi et al., 2018). Similarly, atlas parcellations driven by group may not adequately capture individual variability in brain networks (Li et al., 2022). However, NMF allows for a more flexible and individualized representation of functional networks. In this study, we applied this approach to rs-fMRI data. We conducted an analysis involving FC and employed graph-theory-based network assessments to reveal alterations in the brain associated with CID. Furthermore, we performed correlation analyses to explore potential relationships between neural indicators and clinical manifestations. Our research pertains to uncovering the underlying neural mechanisms of CID. Based on this individualized approach, we hypothesize that (i) CID would show robust alterations in large-scale FNC and graph-theoretic topology when these measures are computed on subject-specific network nodes, and (ii) these alterations would be most prominent in networks implicated in insomnia, including default mode, visual, and attention-related systems. This framing directly targets the added value of individualized mapping by testing whether CID-related effects persist after inter-individual variation in functional topography is explicitly modeled.

Materials and methods

Participants

The rs-fMRI data from 133 study participants were collected at Chengdu Second People’s Hospital (CSPH) between June 2020 and July 2022. The study received approval from the Research Ethics Committee of CSPH (Ethics number 2020021). In this study, there were two categories of participants: CID and GSCs. Prior to rs-fMRI data collection, all participants were informed about the purpose and procedures of the study and voluntarily signed informed consent forms. With the guidance from doctors and nurses, 133 participants completed MRI scans and neuropsychological tests, including the Pittsburgh Sleep Quality Index (PSQI), Self-Rating Depression Scale (SDS), and Self-Rating Anxiety Scale (SAS).

Nine participants (seven with CID and two GSCs) were excluded because their head motion exceeded 2 mm or 2° (see rs-fMRI data preprocessing for details). The sample consisted of 86 individuals with CID (27 males/59 females, mean age = 34.40 years, standard deviation = 11.06) and 38 GSC (9 males/29 females, mean age = 35.11 years, standard deviation = 8.82), totaling 124 participants.

The inclusion criteria for individuals with CID were as follows: (i) meeting the diagnostic criteria for CID as outlined in the 3rd version of the International Classification of Sleep Disorders criteria (Sateia, 2014); (ii) having a PSQI score higher than 7; (iii) refraining from the use of any hypnotic medication for a minimum of 2 weeks before undergoing neuropsychological testing and MRI scanning; and (iv) being between the ages of 18 and 55 years. The GSC group was matched with the CID group in terms of age, gender, and educational level. The inclusion criteria for the healthy control group were as follows: (i) experiencing good sleep quality; (ii) maintaining regular sleep patterns; (iii) having no history of substance misuse, neurological, or psychiatric disorders; and (iv) displaying no abnormalities detected in routine brain imaging examinations. Exclusion criteria were uniformly applied to all participants and included the following: (i) a history of other neuropsychiatric disorders and serious chronic illnesses (e.g. diabetes, heart disease, and cancer); (ii) the presence of other sleep disorders, such as sleep-related breathing disorders (sleep apnea syndrome), central disorders of hypersomnolence, circadian rhythm sleep–wake disorders, sleep-related movement disorders, parasomnia, and hypersomnia; (iii) a history of substance addiction (e.g. drugs, nicotine, alcohol); (iv) the identification of brain lesions or white matter hyperintensities through routine T2-weighted MRI scans; and (v) during fMRI imaging if the subject’s head movement exceeded 2.0 mm or 2.0°.

Clinical assessments

Before fMRI scans, each participant completed neuropsychological tests, including the PSQI, SDS, and SAS. The PSQI, developed by Buysse et al. (1989), was employed to assess the sleep quality of all participants. The SDS, proposed by Zung et al. (1965), was utilized to assess the level of depression, offering a convenient, straightforward, and user-friendly assessment tool. Similarly, the SAS, introduced by Zung (1971), was used to gauge anxiety levels, providing a practical and easy-to-administer evaluation instrument.

rs-fMRI data preprocessing

Before formal data analysis, the head fMRI images in the initial DICOM medical format were converted to NII format. Subsequently, the collected fMRI data were preprocessed using the DPARSFA (https://rfmri.org/DPARSF) toolbox in the MATLAB2014b platform. The preprocessing steps were as follows. (i) The first 10 time points of each participant were removed to allow for participant and gradient magnetic field adaptation and stabilization, minimizing external interference on the subsequent experimental results. (ii) Temporal slice timing correction was applied to the acquired images. Since the brain was scanned interleaved in this experiment, there were significant temporal differences within the same brain region, which were corrected to align at the same time point. (iii) Head motion correction was performed for each participant’s fMRI data to mitigate the potential impact of head motion on image quality during data acquisition. (iv) Normalization was conducted with voxel size set to 3 × 3 × 3 mm3. (v) Spatial smoothing (full width at half-maximum, 6 mm) was applied to reduce registration inaccuracies and enhance signal-to-noise ratio. (vi) Linear drift removal was performed to address thermal noise generated by the machine during data acquisition, resulting in cleaner data. (vii) Band-pass filtering was applied with a range of 0.01–0.1 Hz. Given the significance of low-frequency signals in fMRI, this filtering approach helps focus subsequent analyses on meaningful signal frequencies. (viii) Covariate regression was applied to reduce the effects of Friston-24 head motion parameters, white-matter, cerebrospinal-fluid, and global brain signals. (ix) Head motion artifacts were further addressed to eliminate motion effects caused by head movement.

Definition of 17 personalized functional networks

To facilitate a more intuitive analysis of the functional brain regions of all participants, we employed NMF to construct individual functional network mappings. This approach segmented all voxels of the brain into 17 brain regions based on the Yeo brain parcellation scheme (Yeo et al., 2011). Functional connectivity analyses were then performed between these brain regions. This not only allowed us to visually identify changes in the brain’s functional regions in insomnia patients but also improved computational efficiency. The process of building individual functional brain networks using NMF can be summarized as follows: (i) initialization of group-level networks; (ii) construction of group-level network maps; and (3) Definition of personalized networks. Critically, these individualized network maps served as the node definitions for all subsequent connectome construction, large-scale FNC analyses, and graph-theoretic metrics (See the sections titled Methodology for constructing brain functional connectomes, Large-scale functional connectivity strength differences, and Network attributes)). Thus, every connectivity estimate and topological metric reported below is grounded in subject-specific network topography rather than a single group-level parcellation.

We initially randomly selected 100 samples from the pool of 124 participants. Previous research has demonstrated that the number of randomly selected samples does not affect the results. The time series data of these 100 randomly selected samples were then combined into a matrix with dimensions of 23 000 rows (time points) and 67 541 columns (voxels). Next, we applied the NMF method (Li et al., 2017) to classify functional networks into 17 categories. This large matrix was decomposed into two matrices multiplied together: one matrix represented the group-level network loadings (V matrix) with dimensions of 17 rows and 67 541 columns. Each row of this matrix represented a functional network, and each column represented the loading of a given cortical vertex. The other matrix had dimensions of 17 × 23 000. To enhance robustness, this process was repeated 50 times, with each subsample being randomly selected. Subsequently, spectral clustering was applied to these 50 different group-level network maps to obtain a group-level atlas. We integrated the 50 sets of functional networks into one matrix, which consisted of 850 rows (representing functional networks) and 67 541 columns (representing voxels). Using normalized cut and spectral clustering methods, we partitioned the 850 functional networks into 17 clusters.

We initialized individual-specific network maps using the population consensus atlas and minimized the objective function of the regularized NMF model through an iterative multiplicative updating strategy to achieve convergence. Thus, we created probabilistic (soft) network maps for each individual, providing 17 loadings for each vertex to quantify its membership in the 17 networks. Within each cluster, we selected the most representative functional network that exhibited the highest overall similarity with other functional networks within the same cluster.

Despite each individual showing different network topologies, there were significant similarities among them in certain aspects. Therefore, in the aforementioned steps, we first generated a group-level atlas, which served as the initialization for individualized networks. This approach ensured spatial correspondence among all participants. To mitigate the influence of outliers on the group-level atlas and reduce computational costs, in this study, we randomly selected participant samples for group-level decomposition and repeated the process multiple times. The decomposition results were then aggregated into a reproducible and robust group-level atlas. Subsequently, it was transformed into the form of two matrices based on the number of brain regions divided into 17 clusters, resulting in the final individualized functional networks.

Methodology for constructing brain functional connectomes

In this study, functional connectivity networks were assessed by computing Pearson correlation coefficients between time series. The brain was divided into 17 independent cortical and subcortical regions of interest, with each region representing a network node. Subsequently, extensive functional network topology analysis was conducted, computing functional connections using the Pearson correlation coefficient method. The specific formula for calculating the Pearson correlation coefficient is as follows:

graphic file with name TM0001.gif

When computing node FC, x and y represent two independent nodes, where xi and yi represent the blood oxygen level intensity at the ith time point for voxel x and y, and \bar x and \bar y represent the average blood oxygen level intensity for voxel x and y. The Pearson correlation coefficient ranges from −1 to 1. A value of 0 indicates no correlation, negative values indicate a negative correlation, and positive values indicate a positive correlation. An absolute value of 1 indicates a perfect correlation between variables (1 for perfect positive correlation, −1 for perfect negative correlation). When x and y are the same node, the Pearson coefficient is 1. Consequently, each participant obtains a 17 × 17 node functional connectivity matrix, which is a symmetric matrix with diagonal elements of 1. These functional connectivity matrices are utilized for subsequent statistical and graph theory analyses. Thus, nodes correspond to specific brain regions. After obtaining the node functional connectivity matrices, significant differences between GSCs and CID were analyzed. If significant differences were found, further two-sample t-test were performed to determine differences in connectivity strength between groups, with the significance level also set at P < 0.05 and adjusted for false discovery rate (FDR).

Large-scale functional connectivity strength differences

To investigate the large-scale functional connectivity differences between the two experimental groups, we employed the two-sample t-test. This allowed us to directly compare the connectivity strengths between CID and GSC. The significance level was set at P < 0.05. Additionally, nodal network properties were corrected using the FDR method to account for potential associations between different connectivity measures. This approach enabled a comprehensive examination of the functional connectivity alterations between the two groups while controlling for false positives arising from multiple comparisons.

Network attributes

To investigate the complex brain networks that encompass interconnected functional associations between brain regions, we incorporated small-world properties from graph theory. Small-world organization combines high clustering with short characteristic path length relative to matched random graphs, enabling efficient segregation and integration of information in brain networks (Watts and Strogatz, 1998). This architecture is widely regarded as near-optimal for information processing and is a hallmark of healthy brain function. We utilized the Gretna toolbox (https://www.nitrc.org/projects/gretna/) in Matlab2016b for further analysis. We calculated the small-world properties of large-scale functional networks, which encompass small-worldness (Sigma) >1, normalized clustering coefficient (Gamma) ≫1, and normalized characteristic path length (Lambda) ≈1. We set the step size of network sparsity to 0.01, ranging from 0.1 to 0.4. Consistent with prior rs-fMRI graph studies that quantify small-worldness on binary networks (Wang et al., 2009; Li et al., 2018b) to reduce susceptibility to outliers and noise, we thresholded the functional connectivity matrices to obtain undirected, unweighted adjacency matrices across a range of network densities. Negative correlations in the functional connectivity matrices were converted to their absolute values prior to thresholding, as this approach allows incorporating the strength of anticorrelations while ensuring consistency in network construction. Upon meeting the small-world criterion, we proceeded to calculate the global and nodal topological parameters, including clustering coefficient (Cp), shortest path length (Lp), local efficiency (Eloc), degree centrality (Deg), betweenness centrality (Be), nodal efficiency (Ne), global efficiency (Eglob), assortativity (r) and modularity (Q).

For each network metric, we assessed the differences between groups by comparing their mean values. Significance analysis was performed using two-sample t-tests, and the FDR was applied for multiple comparisons correction.

Statistical analysis

The assumptions of normality and homogeneity of variance for all group comparisons were tested using the Shapiro–Wilk test and Levene’s test, respectively. Data that met these assumptions were analyzed using parametric tests (t-tests). For data that violated these assumptions, non-parametric equivalents (Kruskal–Wallis test followed by Mann–Whitney U tests with Bonferroni correction) were employed. Based on tests for normality and variance homogeneity, group comparisons for age, education, and neuropsychological scale scores (PSQI, SDS, SAS) were conducted using the corresponding parametric or non-parametric tests, while a chi-square test was used for categorical gender data. A significance level of P < 0.05 employed to determine statistical significance.

Finally, we performed correlation analyses using Pearson correlation coefficients to examine the potential relationships between clinical performances and neuroimaging measurements. Significance was determined at P < 0.05 (FDR corrected).

Results

Demographic characteristics and clinical data features

Comparing the CID and GSC groups, there were no statistically significant differences in gender, age, or educational level (P > 0.05). This suggests that chronic insomnia is not significantly associated with factors such as gender, age, or educational level. However, the CID group had significantly higher scores in PSQI, SDS, and SAS compared to the GSC group, all of which were statistically significant (P < 0.001). The demographic characteristics and clinical data of all participants are presented in Table 1.

Table 1.

Demographics and clinical characteristics of the CID and GSC groups.

Variables GSC (n = 38) CID (n = 86) t/χ2 P value
Age (years) 34.39 ± 8.71 35.10 ± 11.00 t = −0.352 0.09
Gender (M/F) 9:29 27:59 χ2 = 0.761a 0.396
Year of education 14.71 ± 3.65 15.47 ± 2.31 t = −1.414 0.565
Duration (years) 6.67 ± 5.76
PSQI 3.39 ± 1.63 12.87 ± 3.22 t = −17.175 <0.001
SDS 38.49 ± 10.36 52.24 ± 11.47 t = −6.334 <0.001
SAS 34.71 ± 8.20 50.17 ± 12.20 t = −7.129 <0.001
Open in a new tab

Data are presented as mean ± SD. aThe P value was obtained by two-tailed Pearson chi-square test. Other P values were obtained by Mann–Whitney U tests.

Abbreviations: CID, chronic insomnia disorder; GSC, good sleeper control; PSQI, Pittsburgh Sleep Quality Index; SDS, Self-Rating Depression Scale; SAS, Self-Rating Anxiety Scale.

Individualized functional networks extracted by NMF

In our study, we applied a spatially regularized NMF method to extract 17 distinct functional networks for each subject. These networks were obtained through an NMF approach and are considered large-scale in nature. We compared these identified networks with the established Yeo’s atlas and successfully classified them accordingly. Specifically, the results can be divided into the default mode network (DMN: DMN-1, DMN-2, DMN-3, and DMN-4), visual network (VIS: VIS-1, VIS-2, VIS-3, and VIS-4), ventral attention network (VAT), somatomotor network (MOT: MOT-1, MOT-2, and MOT-3), frontoparietal network (FPN: FPN-1 and FPN-2), dorsal attention network (DAT), and cerebellum network (CR: CR-1 and CR-2). The group-level functional network and two randomly chosen individual functional networks are displayed in Fig. 1.

Figure 1.

Figure 1

Open in a new tab

Randomly selected CID, GSC, and 17 group-level brain networks.

In Yeo et al. (2011), the 17 functional networks were assigned descriptive labels based on their spatial locations and correspondence with canonical brain regions identified in prior literature. We followed a similar approach by mapping each component to the most spatially overlapping canonical region (Supplementary Excel file 1).

Abnormal large-scale functional connectivity

It is evident that there are abnormal large-scale FNC differences between CID and GSCs, as shown in Fig. 2. When compared to GSCs, CID showed decreased functional connectivity in several areas, including between VAT and DMN-3, MOT-1 and VIS-4, MOT-1 and MOT-3, and VIS-2 and DMN-3. Moreover, CID exhibited higher functional connectivity between VIS-2 and DMN-4, FPN-2 and MOT-2, and MOT-2 and DMN-3 compared to GSCs. These findings suggest distinct functional network patterns between the CID and GSC groups. Notably, these between-group effects were observed on individualized network nodes derived for each subject, suggesting that CID-related large-scale disconnectivity remains detectable even when inter-individual variation in functional network topography is explicitly modeled.

Figure 2.

Figure 2

Open in a new tab

Significant differences in FNC between the CID and GSC groups. *P < 0.05.

Differences in topological properties

Based on FNC, we calculated the topological properties. In both the CID and GSC groups, small-world properties were observed, indicating that the brain networks of both CID and GSCs exhibit small-world organization. Values of Cp, gamma, lambda, Lp, and sigma were compared across different thresholds, with significant differences at certain thresholds marked with asterisks. Except for Cp, which increases with the threshold, the other values of gamma, lambda, Lp, and sigma decrease with increasing thresholds. Furthermore, at the same threshold, the values for CID are consistently higher than those for GSCs. Also, significant group-level differences (P < 0.05) were found in the small-world indices sigma and gamma under the value of the area under the curve (AUC). Specifically, the values of sigma, gamma, lambda, and Cp were all significantly higher in the CID group compared to the GSC group. The Lp value of the CID is nearly identical to that of the GSC, as shown in Fig. 3.

Figure 3.

Figure 3

Open in a new tab

Changes in small-world topology properties between the CID and GSC groups at different thresholds.

Moreover, subjects with CID demonstrated significantly increased AUCs of Eloc and the AUCs of Cp in DMN and increased Deg in the VIS network, compared to GSCs. However, the AUCs of Ne in DMN were significantly smaller in the CID group compared to the GSC group. No other differences in nodal topological properties were observed, as shown in Fig. 4.

Figure 4.

Figure 4

Open in a new tab

Significant differences in the small-world topology networks between the CID and GSC groups. *P < 0.05.

Brain–behavior relationship

In the CID group, there was no significant correlation between the small-world attribute value (average for sparsity 0.1–0.4) of the CID’s whole-brain network and clinical scale scores (PSQI, SDS, SAS). Other brain regions that showed significant differences between the CID and GSC groups did not exhibit any significant correlations (|r|  < 0.1) with clinical scale scores.

Discussion

In this study, we employed rs-fMRI and combined NMF brain decomposition with network topological properties analysis to investigate alterations in large-scale brain functional networks in CID. In contrast to prior studies, we employed an individual-specific functional network approach rather than the commonly used group-level approach for investigating FC. Our results revealed that individual functional network topological properties and individual large-scale FNC could effectively distinguish CID cases. These findings provide novel evidence for neuropsychiatric disorders such as CID. Moreover, our study demonstrates the potential of individual functional mapping as a promising approach to identifying personalized neurobiomarkers for CID. In addition, insights emerging from mapping intrinsic brain connectivity networks offer a potentially mechanistic framework for understanding aspects of human behavior and mental disorders, which aligns with recent advances in systems neuroscience (Yu et al., 2025; Gong et al., 2025). Because all analyses are anchored to individualized network topography, these findings are not contingent on imposing a one-size-fits-all atlas; instead, they provide subject-specific evidence of CID-related network disruption while explicitly accounting for inter-individual variability.

Changes in functional connectivity

After employing the NMF method to partition the brain into global networks, we treated the brain as 17 brain network nodes and conducted functional connectivity matrix analysis using Pearson’s coefficient correlation method. We observed significant differences in the functional connectivity network, particularly in key brain network regions such as the DMN, VIS, and MOT. The DMN, known for its role in self-referential thinking and internal mentation during rest, showed abnormally increased intra-network connectivity in CID patients. This may reflect the excessive rumination and worry commonly reported in insomnia, which have been linked to disrupted DMN function (Marques et al., 2018; Xia et al., 2022). Our findings support previous evidence of an association between sleep disturbance and DMN abnormalities (Ruoff and Rye, 2016), reinforcing the view that DMN dysregulation is a central feature of CID.

The VIS is crucial in processing visual information, and our study found that the connectivity between the VIS and DMN was significantly weakened in CID patients. Dai et al. (2020) previously identified FNC changes between the VIS and DMN. These changes may affect patients' cognitive resource allocation during visual tasks such as reading or driving, thereby impacting daily life. Thus, we can infer that the neurobiological mechanisms underlying CID are not merely attributed to isolated changes in individual brain regions but rather result from complex interactions among multiple typical brain networks. This finding is consistent with our initial hypothesis.

These findings emphasize that chronic insomnia is not only associated with functional changes in individual brain regions but also broadly involves functional imbalances between multiple networks. Specifically, changes in the DMN may be central to the neurobiological alterations of chronic insomnia, reflecting the complexity and multidimensionality of this disorder. Through this comprehensive network analysis, we further confirmed the critical role of the DMN in the pathology of insomnia and revealed possible functional dysregulation between the DMN and visual and motor networks. Our study not only provides new perspectives for understanding the complex neural mechanisms of CID but also suggests potential targets for future therapeutic strategies. Identifying and regulating the functional connectivity between these key networks, particularly the DMN, may become new approaches to alleviate insomnia symptoms. Future research needs to further explore how these network abnormalities develop over time and their response patterns to different therapeutic interventions, providing more precise treatment strategies for clinical practice. Observed DMN and VIS disruptions map onto common CID symptoms such as rumination and attention deficits; this pattern motivates discussion of neuromodulatory targets. Recent randomized neuroimaging work shows that neuronavigated 1 Hz repetitive transcranial magnetic stimulation (rTMS) over the right dorsolateral prefrontal cortex (DLPFC) improves insomnia and mood symptoms and reorganizes DLPFC-centered connectivity, with DLPFC FC changes partially mediating clinical benefit (Gong et al., 2025). These results support the broader therapeutic hypothesis that targeted modulation of hyperconnected hubs could normalize large-scale network dynamics relevant to CID.

Changes in network properties

Consistent with previous findings, we also found that both the CID and GSC groups exhibited economically small-world properties (Xia and He, 2017; Li et al., 2018b). However, the current findings of differences in global network metrics between CID and GSC groups are heterogeneous. Li et al. (2018b) reported no significant difference in small-world properties whereas Huang et al. (2021) found that compared to GSCs, the values of Cp, sigma, Eloc, lambda, and gamma for CID are significantly lower. For CID, Wu et al. (2018) revealed that CID exhibited significantly lower sigma, gamma, Eg, and Eloc, and higher lambda and Lp. Qi et al. (2022) only found that CID reduced the small-world property sigma. Nevertheless, a recent study demonstrated that there were no significant differences in global network measures between CID and GSCs but the CID group exhibited an increased Lp, gamma, lambda, and Eloc, decreased sigma and Eglob and similar Cp in the anatomic brain networks as compared with the GSC group (Lu et al., 2017).

In this study, we found higher sigma, gamma, lambda, Lp, and Cp in CID than GSCs. Additionally, there was no significant difference between CID and GSCs in Eglob. Increased small-world properties suggest decreased brain functional segregation and integration ability in subjects with CID (Wang et al., 2019). However, the majority of previous studies tend to yield results indicating a lower Eglob. Eglob is an excellent indicator of functional integration capacity (Rubinov and Sporns, 2010). CID has a lower Eglob compared to GSCs, indicating that insomnia may exert certain negative effects on the capacity for functional integration, leading to impaired information transmission capability.

While our findings exhibit variations compared to previous studies, there are some similarities. All findings adhere to small-world properties, and significant differences are observed in several key parameters, whether they increase or decrease. This suggests alterations in the brain connectivity of CID. However, the differing results may be attributed to the methods employed in constructing the initial functional connectivity matrices and node definitions. Most previous studies defined brain regions derived from a template, whereas, in our study, large-scale networks were utilized as nodes. This is a rare application of large-scale networks in CID research.

In addition to the significant differences observed in the global network metrics, certain parameters of CID and GSCs also exhibit significant distinctions within the nodal network metrics. Specifically, CID demonstrates significant variations in Cp, Ne, and Eloc for the DMN, showcasing higher Cp and Eloc values and lower Ne values compared to GSCs. Meanwhile, in terms of the parameter Deg, CID displays significantly higher values than GSCs for VIS. Liu et al. (2018) demonstrated that there are anomalous nodal attributes of hub nodes within complex functional networks such as DMN and VIS in CID. The DMN plays a crucial role in processing self-relevant information, encompassing aspects such as episodic memory, social cognition, empathy, mind wandering, theory of mind, and decision-making (Raichle, 2015; Marques et al., 2018; Li et al., 2023). The significant alterations in the DMN largely indicate that chronic insomnia can lead to changes in the brain. Without intervention or treatment, symptoms such as difficulty falling asleep and shallow sleep are likely to persist throughout an individual’s life, potentially triggering additional disorders such as depression and anxiety.

What do we gain from individualized functional networks?

Individualized network mapping is intended to preserve subject-specific functional topography, which is increasingly recognized as meaningful rather than nuisance variability. Connectome fingerprinting studies demonstrate that an individual’s functional connectivity pattern can be distinctive and reliable across sessions, enabling identification and prediction of cognitive traits (Finn et al., 2015; Cai et al., 2021). In case-control settings, using fixed atlas nodes can introduce template-misalignment bias: if network boundaries vary across individuals (or differ systematically between groups), averaging signals within mismatched regions may attenuate true effects and contribute to inconsistent findings. By constructing subject-specific, NMF-derived networks that remain aligned through the shared factorization framework, our study evaluates CID-related FNC and topology directly on individualized nodes. Accordingly, the DMN/VIS/VAT disconnectivity and small-world/nodal alterations reported here reflect CID-related reorganization expressed on each subject’s own network layout, which clarifies the added value of individualized functional networks for CID.

Having established these robust group-level alterations on individualized network nodes, we next asked whether the same individualized features also tracked clinical heterogeneity within CID. Specifically, we examined associations with symptom measures (PSQI, ISI, and sleep duration). However, despite the improved sensitivity for detecting case–control differences, we did not observe significant relationships between personalized network features and symptom severity. In the CID group, the whole-brain small-world attribute showed no association with PSQI, SDS, or SAS, and regions/networks that differed from GSCs likewise did not correlate with these scales. This pattern may indicate that individualized topology contributes more to detecting diagnostic-group effects than to explaining variance in symptoms captured by global questionnaires. It may also reflect limits of our measures and design: PSQI may be too coarse, symptom variance may be restricted, scan and rating times may not align, patients may be heterogeneous, and strict multiple-comparison control can hide small effects. These explanations are not mutually exclusive. Notably, other individualized representations have yielded positive brain–symptom links in independent cohorts; for example, functional connectome gradients predicted PSQI, SDS, and SAS in CID using cross-validation and permutation testing, suggesting that representation choice may influence sensitivity to symptom variance (Wu et al., 2024). This divergence reinforces the need to integrate multivariate, possibly non-linear models and symptom dimensions, and to align imaging with richer state measures and larger samples.

Limitations

There are several limitations to this study. First, the sample size of the GSC group is relatively small, which may have had some impact on the results obtained. Increasing the sample size in future studies could enhance the reliability and persuasiveness of the findings. Second, some related research has indicated that brain networks such as the DMN change disorders such as depression or anxiety, which often co-occur with insomnia. In this study, aside from the SDS and SAS scales, we did not further investigate the correlation of these mental disorders to control for variables. Future research can explore this aspect in greater depth. Third, although we adopted a repeated subsampling (50 iterations) and spectral clustering procedure to mitigate variability from the probabilistic decomposition and group-level initialization in NMF, the stability of the individualized network mapping approach under different random initializations or across independent datasets was not systematically assessed. Future work should evaluate reproducibility metrics, such as spatial similarity indices, on multi-site datasets or repeated scans, to establish the generalizability of this method in broader research and clinical contexts.

Conclusion

Utilizing individual large-scale FNC and graph theory analysis methods, our study revealed multiple abnormalities in the topological parameters and functional connectivity of the brain network in CID, confirming extensive impairment of CID’s brain functional network. To the best of our knowledge, this study represents one of the early instances both nationally and internationally where the individual-level functional network mapping approach has been applied to research on CID. Our study has provided objective neuroimaging evidence for the neurobiological mechanisms of CID, advancing our understanding of this disorder. Additionally, our research has unveiled topological and functional abnormalities in the brains of individuals with chronic insomnia, offering a novel perspective on understanding the clinical symptoms and neurobiological underpinnings of chronic insomnia. By constructing and analyzing individualized large-scale networks rather than relying solely on group-level parcellations, this study provides subject-specific evidence of disrupted connectivity and topology in CID and supports personalized network approaches as a promising direction.

Supplementary Material

kkag001_Supplemental_File
kkag001_supplemental_file.zip (8.6KB, zip)

Acknowledgements

This work was supported by Sichuan Science and Technology Program (Grant No.2024YFHZ0357, 2024ZYD0136, 2026YFHZ0202), National Natural Science Foundation of China (No.82371479, 82001803).

Contributor Information

Xiaoxuan Li, College of Electrical Engineering, Sichuan University, Chengdu 610065, China.

Yiran Zhai, College of Electrical Engineering, Sichuan University, Chengdu 610065, China.

Tianwei Qin, College of Electrical Engineering, Sichuan University, Chengdu 610065, China.

Zhiwei Zhang, College of Electrical Engineering, Sichuan University, Chengdu 610065, China.

Zixi Zhao, College of Oxford Brookes, Chengdu University of Technology, Chengdu 610059, China.

Hanbin Deng, Sichuan Institute of Computer Sciences, Chengdu 610041, China.

Siqi Yang, School of Cybersecurity, Chengdu University of Information Technology, Chengdu 610225, China.

Jiang Zhang, College of Electrical Engineering, Sichuan University, Chengdu 610065, China.

Liang Gong, Department of Neurology, West China School of Medicine, Sichuan University, Sichuan University affiliated Chengdu Second People’s Hospital., Chengdu 610017, China.

Author contributions

Xiaoxuan Li (Conceptualization, Data curation, Methodology, Validation, Visualization, Writing – original draft, Writing – review & editing), Yiran Zhai (Data curation, Formal Analysis, Resources, Supervision, Validation, Writing – review & editing), Tianwei Qin (Data curation, Formal Analysis, Validation, Writing – original draft, Writing – review & editing), Zhiwei Zhang (Data curation, Formal Analysis, Visualization, Writing – review & editing), Hanbin Deng (Conceptualization, Investigation, Methodology, Supervision, Writing – review & editing), Siqi Yang (Data curation, Investigation, Supervision, Validation), Jiang Zhang (Conceptualization, Project administration, Resources, Software, Supervision, Writing – original draft), Liang Gong (Data curation, Funding acquisition, Investigation, Project administration, Resources, Writing – review & editing), and Zixi Zhao (Formal Analysis)

Conflicts of interests

The authors declare that no competing interests exist.

References

  1. Balay-Dustrude  E, Shenoi  S (2023) Current validated clinical and patient reported disease outcome measures in juvenile idiopathic arthritis. Open Access Rheumatol: Res Rev. 15:189–206. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bi  X-A, Sun  Q, Zhao  J  et al. (2018) Non-linear ICA analysis of resting-State fMRI in mild cognitive impairment. Front Neurosci. 12:413. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Buysse  DJ, Reynolds  CF, Monk  TH  et al. (1989) The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatry Res. 28:193–213. [DOI] [PubMed] [Google Scholar]
  4. Cai  B, Zhang  G, Zhang  A  et al. (2021) Functional connectome fingerprinting: identifying individuals and predicting cognitive functions via autoencoder. Hum Brain Mapp. 42:2691–705. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Chaput  J-P, Janssen  I, Sampasa-Kanyinga  H  et al. (2023) Economic burden of insomnia symptoms in Canada. Sleep Health. 9:185–9. [DOI] [PubMed] [Google Scholar]
  6. Cheng  Y, Xue  T, Dong  F  et al. (2022) Abnormal functional connectivity of the salience network in insomnia. Brain Imaging Behav. 16:930–8. [DOI] [PubMed] [Google Scholar]
  7. Crossley  NA, Mechelli  A, Scott  J  et al. (2014) The hubs of the human connectome are generally implicated in the anatomy of brain disorders. Brain. 137:2382–95. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Cui  Z, Li  H, Xia  CH  et al. (2020) Individual variation in functional topography of association networks in youth. Neuron. 106:340–353 e8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Dai  X-J, Liu  B-X, Ai  S  et al. (2020) Altered inter-hemispheric communication of default-mode and visual networks underlie etiology of primary insomnia : altered inter-hemispheric communication underlie etiology of insomnia. Brain Imaging Behav. 14:1430–44. [DOI] [PubMed] [Google Scholar]
  10. Finn  ES, Shen  X, Scheinost  D  et al. (2015) Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity. Nat Neurosci. 18:1664–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Gong  L, Yang  X, He  Y  et al. (2025) Precision targeting of right dorsolateral prefrontal cortex with neuronavigated rTMS alleviates chronic insomnia via functional connectivity reorganization: a randomized neuroimaging trial. Neuroimage Clin. 47:103815. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Hertenstein  E, Feige  B, Gmeiner  T  et al. (2019) Insomnia as a predictor of mental disorders: a systematic review and meta-analysis. Sleep Med Rev. 43:96–105. [DOI] [PubMed] [Google Scholar]
  13. Huang  W, Li  Z, Wu  S  et al. (2021) [Small-world network of patients with primary insomnia: a resting-state functional magnetic resonance imaging study]. Nan Fang Yi Ke Da Xue Xue Bao. 41:424–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Langs  G, Wang  D, Golland  P  et al. (2016) Identifying shared brain networks in individuals by decoupling functional and anatomical variability. Cereb Cortex. 26:4004–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Laumann  TO, Gordon  EM, Adeyemo  B  et al. (2015) Functional system and areal organization of a highly sampled individual Human brain. Neuron. 87:657–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Lee  DD, Seung  HS (1999) Learning the parts of objects by non-negative matrix factorization. Nature. 401:788–91. [DOI] [PubMed] [Google Scholar]
  17. Li  G, Zhang  X, Zhang  J  et al. (2018a) Magnetic resonance study on the brain structure and resting-state brain functional connectivity in primary insomnia patients. Medicine (Baltimore). 97:e11944. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Li  H, Satterthwaite  TD, Fan  Y (2017) Large-scale sparse functional networks from resting state fMRI. Neuroimage. 156:1–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Li  Y, Liu  A, Fu  X  et al. (2022) Atlas-guided parcellation: individualized functionally-homogenous parcellation in cerebral cortex. Comput Biol Med. 150:106078. [DOI] [PubMed] [Google Scholar]
  20. Li  Y, Yu  X, Ma  Y  et al. (2023) Neural signatures of default mode network in major depression disorder after electroconvulsive therapy. Cereb Cortex. 33:3840–52. [DOI] [PubMed] [Google Scholar]
  21. Li  Z, Chen  R, Guan  M  et al. (2018b) Disrupted brain network topology in chronic insomnia disorder: a resting-state fMRI study. Neuroimage Clin. 18:178–85. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Liu  X, Zheng  J, Liu  B-X  et al. (2018) Altered connection properties of important network hubs may be neural risk factors for individuals with primary insomnia. Sci Rep. 8:5891. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Lu  F-M, Dai  J, Couto  TA  et al. (2017) Diffusion tensor imaging tractography reveals disrupted white matter structural connectivity network in healthy adults with insomnia symptoms. Front Hum Neurosci. 11:583. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Lunkova  E, Guberman  GI, Ptito  A  et al. (2021) Noninvasive magnetic resonance imaging techniques in mild traumatic brain injury research and diagnosis. Hum Brain Mapp. 42:5477–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Marques  DR, Gomes  AA, Caetano  G  et al. (2018) Insomnia disorder and Brain’s default-mode network. Curr Neurol Neurosci Rep. 18:45. [DOI] [PubMed] [Google Scholar]
  26. Mayer  G, Happe  S, Evers  S  et al. (2021) Insomnia in neurological diseases. Neurol Res Pract. 3:15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Mueller  S, Wang  D, Fox  MD  et al. (2013) Individual variability in functional connectivity architecture of the human brain. Neuron. 77:586–95. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. National Institutes of Health . (2005) National Institutes of Health State of the Science Conference statement on Manifestations and management of chronic insomnia in adults, June 13-15, 2005. Sleep. 28:1049–57. [DOI] [PubMed] [Google Scholar]
  29. Nie  X, Shao  Y, Liu  SY  et al. (2015) Functional connectivity of paired default mode network subregions in primary insomnia. Neuropsychiatr Dis Treat. 11:3085–93. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Qi  S, Zhang  Y, Li  X  et al. (2022) Improved functional organization in patients with primary insomnia after individually-targeted transcranial magnetic stimulation. Front Neurosci. 16:859440. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Raichle  ME (2015) The brain’s default mode network. Annu Rev Neurosci. 38:433–47. [DOI] [PubMed] [Google Scholar]
  32. Rubinov  M, Sporns  O (2010) Complex network measures of brain connectivity: uses and interpretations. Neuroimage. 52:1059–69. [DOI] [PubMed] [Google Scholar]
  33. Ruoff  C, Rye  D (2016) The ICSD-3 and DSM-5 guidelines for diagnosing narcolepsy: clinical relevance and practicality. Curr Med Res Opin. 32:1611–22. [DOI] [PubMed] [Google Scholar]
  34. Sateia  MJ (2014) International classification of sleep disorders-third edition: highlights and modifications. Chest. 146:1387–94. [DOI] [PubMed] [Google Scholar]
  35. Seeley  WW, Menon  V, Schatzberg  AF  et al. (2007) Dissociable intrinsic connectivity networks for salience processing and executive control. J Neurosci. 27:2349–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Spiegelhalder  K, Regen  W, Baglioni  C  et al. (2013) Neuroimaging studies in insomnia. Curr Psychiatry Rep. 15:405. [DOI] [PubMed] [Google Scholar]
  37. Streatfeild  J, Smith  J, Mansfield  D  et al. (2021) The social and economic cost of sleep disorders. Sleep. 44:zsab132. [DOI] [PubMed] [Google Scholar]
  38. Sun  H, Xu  D, Sun  Q  et al. (2024) Functional connectivity analyses of individual hippocampal subregions in major depressive disorder with electroconvulsive therapy. Psychoradiology. 4:kkae030. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Tursunboyev  M, Yaxyayev  M, Turayev  B (2025) Clinical psychopathological features of sleep disorders. Modern Sci Res. 4:130–9. [Google Scholar]
  40. Wang  D, Buckner  RL, Fox  MD  et al. (2015) Parcellating cortical functional networks in individuals. Nat Neurosci. 18:1853–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Wang  J, Wang  L, Zang  Y  et al. (2009) Parcellation-dependent small-world brain functional networks: a resting-state fMRI study. Hum Brain Mapp. 30:1511–23. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Wang  R, Lin  P, Liu  M  et al. (2019) Hierarchical connectome modes and critical state jointly maximize human brain functional diversity. Phys Rev Lett. 123:038301. [DOI] [PubMed] [Google Scholar]
  43. Watts  DJ, Strogatz  SH (1998) Collective dynamics of ‘small-world’ networks. Nature. 393:440–2. [DOI] [PubMed] [Google Scholar]
  44. Wu  J, Yang  J, Yuan  Z  et al. (2024) Functional connectome gradient predicts clinical symptoms of chronic insomnia disorder. Prog Neuropsychopharmacol Biol Psychiatry. 135:111120. [DOI] [PubMed] [Google Scholar]
  45. Wu  Y, Liu  M, Zeng  S  et al. (2018) Abnormal topology of the structural connectome in the limbic Cortico-basal-ganglia circuit and default-mode network among primary insomnia patients. Front Neurosci. 12:860. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Xia  G, Hu  Y, Chai  F  et al. (2022) Abnormalities of the default mode network functional connectivity in patients with insomnia disorder. Contrast Media Mol Imaging. 2022:9197858. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Xia  M, He  Y (2017) Functional connectomics from a “big data” perspective. Neuroimage. 160:152–67. [DOI] [PubMed] [Google Scholar]
  48. Yeo  BT, Krienen  FM, Sepulcre  J  et al. (2011) The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol. 106:1125–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Yu  L, Gong  L, Chen  X  et al. (2025) Exploratory GABAa-informed control network modulates hyperarousal brain dynamics in chronic insomnia. Commun Biol. 8:991. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Zhang  J, Qin  T, Sun  H  et al. (2025) Delineating specific structural and functional patterns for neurophysiological biotypes of childhood maltreatment. Cereb Cortex. 35. [DOI] [PubMed] [Google Scholar]
  51. Zhang  J, Zhang  Z, Sun  H  et al. (2024) Personalized functional network mapping for autism spectrum disorder and attention-deficit/hyperactivity disorder. Transl Psychiatry. 14:92. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Zhang  J, Zhao  T, Zhang  J  et al. (2022) Prediction of childhood maltreatment and subtypes with personalized functional connectome of large-scale brain networks. Hum Brain Mapp. 43:4710–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Zung  WW (1971) A rating instrument for anxiety disorders. Psychosomatics. 12:371–9. [DOI] [PubMed] [Google Scholar]
  54. Zung  WWK (1965) Self-rating depression scale in an outpatient clinic. Further validation of the SDS. Arch Gen Psychiatry. 13:508–15. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

kkag001_Supplemental_File
kkag001_supplemental_file.zip (8.6KB, zip)

Articles from Psychoradiology are provided here courtesy of Oxford University Press

RESOURCES