Clinically “asymptomatic” carotid stenosis, even without stroke, is associated with cognitive impairment (de Weerd et al., 2014; Lal et al., 2017; Lazar et al., 2021), which is generally characterized by reduced processing speed and learning/memory capacity (Martinić-Popović et al., 2009; Lazar et al., 2021). As such, cognitive status has been suggested as an indicator of disease progression and efficacy monitoring in these patients (Lattanzi et al., 2018). However, the neuroanatomical basis underlying cognitive impairment remains largely unclear.

Neuroimaging studies have revealed notable structural alterations in asymptomatic carotid stenosis. Morphological analyses, for example, reveal asymmetrical (Gao et al., 2021b) and stenosis-ipsilateral dominated gray matter (GM) atrophies (Avelar et al., 2015), cortical thinning in the Perisylvian regions (Gao et al., 2021a), and accelerated aging changes (Alhusaini et al., 2018). Lesion mapping has reported prominent subcortical microinfarcts and white matter hyperintensity (WMH; Kandiah et al., 2014; van Veluw et al., 2017). Structural imaging also identifies reduced white matter integrity and fiber density in such patients (Cheng et al., 1970; Meng et al., 2017; Gao et al., 2019). Moreover, these patients are reported to show compromised hemodynamics (Silvestrini et al., 2009; Balucani et al., 2012), glymphatic dysfunction and enlarged perivascular spaces (Liu et al., 2021). These lines of evidence suggest that extensive morphological changes have occurred in these patients, and morphometric network analysis based upon may provide rich information for understanding the neuroanatomical basis of cognitive impairment and predicting long-term risks.

A multivariate methodology is the structural covariance network analysis (Alexander-Bloch et al., 2013), which assumes that morphological measures [e.g., cortical thickness (CT)] of one brain region and another structurally and functionally connected one co-vary, forming different communities. These co-variations are underpinned by gene co-expression (Romero-Garcia et al., 2018), systematically change across lifespan development (Montembeault et al., 2012; Kuo et al., 2020), and are sensitive to the early stages of neuropsychiatric diseases (e.g., Coppen et al., 2016; Liu et al., 2019; Qing et al., 2021). For example, normal aging is characterized by a decrease in GM covariance connectivity in higher-order associative regions, especially those involved in the semantic, executive control, and default mode functions (Montembeault et al., 2012). However, most previous studies of morphometric networks are based on group-level covariance across subjects. While this analysis can reveal brain network architecture, it cannot derive individual-level network measures, which greatly reduces its clinical utility for individual patients.

Consequently, individualized single-subject morphometric network analyses are emerging as a new area of research (Tijms et al., 2012; Li et al., 2017; Xue et al., 2022). They can provide measures similar to those derived from functional connectivity with functional MRI or structural connectivity with diffusion MRI tractography, holding promise for clinical settings. The construction of individual morphometric network is generally by estimating the interregional/areal relationships, e.g., probability distributions, distances, and similarities, of morphological features (e.g., GM volume, CT, cortical gyrification, sulcus depth, and cortical complexity; Tijms et al., 2012; Wang et al., 2016; Li et al., 2017; Raamana and Strother, 2018). The pairwise estimates of morphological features are termed edges in network analysis. The edges and their inter-edge relationships are called edge-centered brain connectivity, which is a new direction in human brain connectomics.

Asymptomatic carotid stenosis is a condition characterized by potential chronic hypoperfusion, microembolism, and hemodynamic burdens caused by internal carotid stenosis (Cheng et al., 1970; Marshall et al., 2017; Wang et al., 2017). This condition primarily affects the Perisylvian cortical reorganization and topographically involves somatosensory/motor, semantic, silent processing, and frontoparietal systems, as well as cognitive functions such as processing speed and semantic memory (Avelar et al., 2015; Gao et al., 2019; Nickel et al., 2019; Gao et al., 2021b). Thus, an investigation of the morphometric network may provide us with a better understanding of the neuroanatomical basis of cognitive impairment in carotid stenosis. Network edges derived from interareal histogram weights form rich connectivity information, and promote the identification of the edges that closely relate to cognitive impairment (Raamana et al., 2015; Wang et al., 2016, 2018; Raamana and Strother, 2020; Li et al., 2021; Peng et al., 2022). However, it is unknown to what extent asymptomatic carotid stenosis may affect the edges and network architecture that are closely associated with cognitive impairment and serve as potential imaging markers.

In this study, we used a novel individual approach to construct single-subject CT networks and contrasted unilateral (>70%) severe asymptomatic carotid stenosis (SACS) patients with demographically and comorbidities-matched healthy controls (HC). We estimated edge-centric structural connectivity (eSC) derived from the pairwise Manhattan distances of regional CT histograms. We expected that the SACS showed abnormal eSC in the Perisylvian and language-related regions and that this eSC was associated with cognitive impairment in SACS patients. Furthermore, large-scale structural networks reconstructed from the eSC would exhibit lower network segregation and efficiency globally.

We estimated the Manhattan distances between pairwise CT histograms to compute eSC and construct subject-wise morphological brain networks. This analysis showed that the SACS patients had longer interhemispheric but shorter intrahemispheric Manhattan distances seeding from left lateral temporal regions, corresponding to lower interhemispheric but higher intrahemispheric connectivity. Further network analysis showed that the SACS patients had decreased system segregation paralleling with WMH burden and recall memory. Lastly, we applied a recently developed NBS approach to get feature weights for eSC, and subgraph features that predicted WMH load and cognitive tests. These results showed that SACS had disrupted structural connectivity, altered eSC measured by Manhattan distances, and pronounced lower system segregation. Further, NBS analysis yielded good feature classification performance. These findings suggest that subject-wise eSC has the potential to identify high-risk individuals while also providing insights into the morphological connectivity basis of cognitive impairment and accelerated aging in advanced asymptomatic carotid stenosis.

This study identified that patients with SACS exhibited altered pairwise CT co-vary patterns with increased interhemispheric but decreased intrahemispheric Manhattan distances seeding from the left lateral temporal regions. This result is consistent with prior reports on atrophy and dysconnectivity in the same regions (e.g., Cheng et al., 1970; Lin et al., 2014; Avelar et al., 2015; Marshall et al., 2017; Tani et al., 2018; Nickel et al., 2019; Gao et al., 2021b), and also generally consistent with previous reports using fMRI functional connectivity or diffusion tensor imaging SC (Lin et al., 2014; Huang et al., 2018; Gao et al., 2019; He et al., 2020). For example, earlier results from resting-state functional connectivity show that SACS has reduced interhemispheric connectivity, especially in the frontoparietal network ipsilateral to stenosis (Cheng et al., 1970). Our prior study also has identified lower interhemispheric functional connectivity located in the Perisylvian regions, spanning across somatomotor, salience, and dorsal attention networks (Gao et al., 2019). We interpret this finding as altered morphological co-variations dominated by the lateral temporal regions. Notably, the SACS exhibited shorter Manhattan distances and thus a higher eSC. This is unusual since neurodegenerative diseases are often associated with decreased connectivity globally. On the one hand, this phenomenon may reflect pairwise synchronized/co-vary atrophy patterns, as much of previous evidence has shown that these regions have thinner cortex in SACS (Avelar et al., 2015; Marshall et al., 2017; Nickel et al., 2019; Gao et al., 2021a), and this synchronized shrinkage could lead to an increased similarity. This speculation requires further verification through computational modeling. On the other hand, this may reflect compensation in the preclinical stage of the disease, as many neurodegenerative diseases have shown a compensatory increase in the early stages.

It is also possible that in SACS patients, the long-range connections that maintain interhemispheric interactions and support advanced cognitive function are impaired, while the relatively short-range intrahemispheric connections that are essential for functional specification are enhanced. Neuroimaging studies have established that interhemispheric integration and intrahemispheric specialization are important underpinnings of an individual’s higher cognitive abilities (e.g., Buckner and Krienen, 2013).

Another novel finding of our study is that the SACS patients exhibited decreased system segregation and comparable system integration as compared to the controls. Human functional imaging has shown that system segregation decreases with normal aging (Chan et al., 2014) and supports cognitive resilience in Alzheimer’s disease (Ewers et al., 2021), and this reduction can be improved through visual speed processing training (Chen et al., 2021). This suggests that the community structure of SACS blurs system and network boundaries, which may reflect a compensatory mechanism under the condition of chronic hypoperfusion and compromised hemodynamics. Such phenomenology has also been suggested as decreased within-system and increased between-system functional connectivity in mild cognitive impairment and Alzheimer’s disease (Vipin et al., 2018).

Limitations. First, as a small sample study, we thus repeated the analyses using different smoothing kernels and atlases to overcome the potential bias; future large-sample studies are needed to address this issue and promote deep learning-based prediction of high-risk individuals, with emphasis on dementia and stroke, to inform clinical decisions for surgical intervention in such patients. Second, since morphological networks and functional connectivity networks have apparently different community structures, the network segregation and integration based on their calculations show different patterns, which is also an active research direction. Third, longitudinal data can provide information on disease progression and prediction of high-risk individuals, together with other clinically commonly used imaging modalities for silent lesion detection, which can lead to better diagnosis and prognosis.

Our results show that SACS exhibits abnormal subject-wise structural connectivity and a less-optimized trade-off between physical cost and network segregation, suggesting asymmetric and/or synchronous cortical atrophy under potential chronic hypoperfusion and disordered hemodynamic pressure. These results also provide a reference and perspective for a future large sample identifying high-risk carotid stenosis individuals.

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding authors.

The studies involving human participants were reviewed and approved by the Medical Ethics Committee of Zhongnan Hospital of Wuhan University. The patients/participants provided their written informed consent to participate in this study.

JR and DX analyzed the data and drafted the manuscript. LG supervised the neuroimaging analyses. HM, XZ, and CF collected the data. JR, XZ, JL, and YX contributed to the data analyses and wrote the manuscript. LG and HX conceived the idea and contributed to the manuscript editing. All authors contributed to the article and approved the submitted version.

This study was supported by the National Natural Science Foundation of China (Grant No. 82001799 to LG) and was partially supported by the Shenzhen Science and Technology Innovation Commission (JCYJ20200109144801736), National Social Science Foundation of China (Nos. 20 & ZD058), and National Natural Science Foundation of China (No. 72031009).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.