The included literature was retrieved from the Core Collection of the Web of Science, spanning from January 1, 2000, to August 31, 2023. The retrieval strategy involved identifying theme words related to neuroimaging and VCI. The specific retrieval formula is shown in Supplementary Table 1, and only articles and reviews pertinent to the research topic will be included for subsequent analysis. Furthermore, publications were restricted to those in English. After comprehensive searching and screening (refer to Figure 1), a total of 1,024 publications, comprising 919 articles and 105 reviews, were selected for this bibliometric analysis. Subsequently, we exported the details containing complete records and cited references in the form of plain text files.

VOSviewer V1.6.19 and CiteSpace V6.2.R3, both Java-based programs, were useful tools for performing bibliometric and visualized analyses. VOSviewer can extract crucial information from included articles to establish collaborative, co-occurrence, and co-citation networks. It calculates similarity matrices by analyzing co-occurrence matrices, subsequently employing the VOS mapping technique to generate a distance-based visualization map (van Eck and Waltman, 2010; Qu et al., 2024). In this study, VOSviewer was applied to illustrate and explore collaborative relationships among authors, countries, and organizations, as well as the co-occurrence networks of keywords and the visualization networks of journals and co-cited journals. Node size denotes the frequency of occurrence, while the thickness of links between nodes depicts connection strength, with nodes automatically clustered and represented in distinct colors. The total link strength is calculated to manifest the cumulative strength of links between two nodes in a visualization network map. In addition, time factors are considered in the overlay visualization analysis of keywords. Nodes of different colors represent different years of keyword occurrences, with each node’s color determined by the average year of the keyword’s appearance.

We employed CiteSpace (Chen, 2006) for clustering analyses of keywords and references, which facilitated a deeper understanding of the research hotspots in VCI neuroimaging. Our analysis was performed with the following parameters: time slicing (2000–2023), 1 year per slice, link strength (cosine), the selection of each slice (g-index with k = 25 as the scale factor), and pruning (pathfinder and pruning the merged network). Clustering analysis utilized the log-likelihood ratio (LLR) algorithm, with each cluster labeled and sorted to elucidate research status and hotspots. A Modularity Q value exceeding 0.3 indicates a significant clustering structure, while a weighted mean silhouette > 0.5 signifies the reliability of the clustering results. Moreover, the identification of keywords and references with strong citation bursts facilitates predicting emerging trends and research frontiers.

A total of 1,024 publications spanning from January 2000 to August 2023 were included in this bibliometric analysis, consisting of 919 articles (89.75%) and 105 reviews (10.25%). As depicted in Figure 2, despite certain fluctuations, the number of publications on VCI neuroimaging shows an overall upward trend. Notably, marked increases were observed in 2014 and 2016, and 2021 was the peak year with 80 publications. These findings indicate the growing attention researchers are devoting to VCI neuroimaging, and further studies are required to advance its development.

The 1,024 retrieved literature involved 5,404 authors. According to the scholar Price’s law (Price, 1963), the minimum publication volume for core authors is m = 0.749 × nmax, where n_max represents the number of published articles conducted by the most prolific author (n_max = 50 here). Hence, the core authors should have released at least six papers. By the calculation of VOSviewer, there are 110 core authors, accounting for 2.04% of the total number of authors. Na, Duk L. emerged as the most prolific author with 50 studies and 1,682 citations. His research concentrated on the role of white matter hyperintensity in cognitive decline, the impact of cerebral amyloid burden on blood viscosity and cognition in subcortical VCI, as well as hippocampal changes, cortical thinning, and microbleeds in subcortical VaD (Seo et al., 2007; Kim et al., 2012, 2015, 2020; Lee et al., 2014; Noh et al., 2014). As listed in Table 1, Scheltens, Philip received the highest average number of citations, manifesting the influence and recognition of his articles among scholars. To facilitate a more intuitive observation of the collaborative relationships of the core authors, VOSviewer was utilized for network visualization analysis. Figure 3 shows that relatively stable cooperative groups have been formed among the core authors.

Analyzing information extracted from publications regarding countries and institutions can help understand their research contributions and collaborative relationships in VCI neuroimaging. Among the 1,024 retrieved articles, 60 countries and 1,431 institutions were involved. Table 2 indicates the USA as the most prolific country, contributing 239 studies, followed by China (212 studies) and England (119 studies). What’s more, the total link strength reflects the collaboration intensity between nations and organizations. The Netherlands exhibits the highest link strength (221), manifesting that the Netherlands has maintained closed connections with other countries in VCI neuroimaging research. Furthermore, Figure 4A displays a cooperation network map of countries with more than five publications. Regarding institutions, Sungkyunkwan University led in published articles (57 papers), followed by Vrije Universiteit Amsterdam (39 papers) and Capital Medical University (29 papers). Remarkably, Sungkyunkwan University also had the highest link strength (239). As shown in Figure 4B, different clusters and cooperation networks have been formed. Intriguingly, the majority of institutions on the top ten list were universities rather than hospital radiology departments.

Table 3 lists the top ten journals in VCI neuroimaging, with the Journal of the Neurological Sciences leading with 50 articles, followed closely by Stroke (46 articles) and Neurology (45 articles). Notably, the 2022 impact factor (IF) of these top ten journals ranged from 2.4 to 11.1, with four being classified as Q1 in the JCR partition. A network visualization map of journals with more than five papers was conducted by VOSviewer and displayed in Figure 5A. In terms of co-cited journals, Neurology, Stroke, and Neuroimage emerged as the most frequently cited journals with 4,637, 3,369, and 1,281 citations, respectively. This visualization map is depicted in Figure 5B, where each node indicates a co-cited journal, categorized into different clusters distinguished by colors. Additionally, most of the top ten co-cited journals belong to the Q1 partition.

Keywords are concise summaries of an article and facilitate readers grasping the core content in a short time. As detailed in Table 4, the most frequently appearing keywords are “dementia” (840 times), also presenting the highest total link strength (5,873), followed by “cognitive impairment” (424 times), “magnetic resonance imaging” (264 times), “small vessel disease” (216 times), and “stroke” (191 times). Figure 6A shows the clustering of keywords, with the largest red cluster mainly containing dementia, diagnostic criteria, and PET, while the green cluster primarily encompassing small vessel diseases, stroke, and white matter hyperintensity. Additionally, the purple cluster comprises functional connectivity, cortex, and diffusion tensor imaging, whereas the yellow clustering mainly includes cognitive impairment, brain, and pathology. In Figure 6B, the purple and blue nodes appear earlier and mainly encompass keywords like diagnostic criteria, dementia, magnetic resonance imaging, and stroke. Conversely, the green and yellow nodes emerge relatively later, involving small vessel diseases, functional connectivity, integrity, and diffusion tensor imaging, which also provide references for predicting future research directions. The timeline map of keywords (Figure 7) depicts the development context and evolution trend of research topics. Prior to 2010, the study mainly focused on cerebral blood flow, stroke, white matter changes, and Binswanger’s disease. There was an extensive exploration of neuroimaging techniques, including artistic spin labeling magnetic resonance, 7T MRI, diffusion tensor imaging, and ¹⁸F-FDG-PET following 2010.

Keyword bursts indicate a rapid increase in the frequency of keyword occurrences over a period of time. It offers insights into research hotspots and emerging frontiers within a specific field. The significance of keywords is demonstrated through their burst strength. As presented in Figure 8, the top three keywords with the strongest burst strengths are “small vessel disease” (2017–2023), “criteria” (2000–2008), and “diagnosis” (2000–2006). Keywords with the latest citation burst, including small vessel disease, connectivity, biomarkers, and lacunar stroke, hold potential for predicting future research directions in VCI neuroimaging.

Co-citation analysis facilitates the identification of a knowledge base and influential publications, allowing for further exploration of the research field. As listed in Table 5, the “Diagnostic and statistical manual of mental disorders” issued by the American Psychiatric Association in 1994 had the highest co-citation count with 72 times and the largest centrality (0.42). It was primarily concerning the diagnostic criteria of VaD. The second most frequently co-cited reference, exhibiting the strongest burst strength (see Figure 9B), was about the neuroimaging standard of small vessel disease (SVD). This article comprehensively analyzes and standardizes the definitions and terminologies of recent small subcortical infarcts, lacune and white matter hyperintensity of presumed vascular origin, perivascular space, cerebral microbleed, and brain atrophy. Moreover, it provides suggestions for the selection of neuroimaging approaches, imaging analysis, and reporting standards for SVD (Wardlaw et al., 2013b).

Clustering analysis of references using CiteSpace yielded 71 clusters, with the top fifteen clusters depicted in Figure 9A. The modularity Q is 0.8421, and the weighted mean silhouette is 0.9383, indicating that the clusters are reasonable and convincing. Clusters #0 (diffusion-weighted imaging), #5 (magnetic resonance imaging), #9 (CT scan), and #13 (SPECT) are important neuroimaging methods concerning VCI. Besides, biomarkers, cerebral amyloid angiopathy, and cerebral blood flow are also considered research hotspots. References with a strong citation burst indicate significant attention and recognition from researchers within a specific timeframe. Gorelick PB’s study obtained the second-greatest citation burst strength (Figure 9B) and ranks third in co-citation counts. This article systematically synthesized the existing evidence, identified the current knowledge gap, and offered recommendations according to standard AHA criteria. It comprehensively covers definitions, diagnostic criteria, neuropathological and neuroimaging factors, cardiovascular risk factors, as well as treatment and prevention strategies related to VCI (Gorelick et al., 2011).

With the continuous advancement in neuroimaging technologies, there is a growing number of studies exploring VCI neuroimaging. Quantitative analysis of papers included from 2000 to 2023 revealed an overall upward trend with some fluctuations in annual publications concerning VCI neuroimaging. This trend manifests an increasing scholarly focus on utilizing neuroimaging modalities to study VCI. Several stable groups of core authors have formed through the analysis of authors. Scheltens, Philip led the largest red cluster, while Na, Duk, L. and Kim, Sung Tae headed the blue and green clusters, respectively. Notably, the USA and China stand out as the most prolific contributors to publications concerning VCI neuroimaging. The Netherlands cooperated closely with other countries in different clusters, such as the USA, Germany, England, and Singapore. Nevertheless, there remains a need to bolster international collaboration for broader academic exchanges. Sungkyunkwan University was the most frequently published institution on the topic of VCI neuroimaging. Stroke and Neurology occupy central positions in the visualization map of journals and co-cited journals, and they are also ranked among the top three in publication counts and citations. This emphasizes their substantial role in advancing VCI neuroimaging research. Among the top ten journals, four of them are classified into the Q1 partition, and the Journal of Neurology, Neurosurgery and Psychiatry has the highest IF of 11.1. Moreover, eight of the top ten co-cited journals belong to the Q1 quartile, among which Lancet Neurology has the highest IF of 48. These journals may also provide options for researchers submitting manuscripts on VCI neuroimaging.

Through the clustering analysis and citation burst of keywords and references, we can capture the research hotspots within the field, which will be discussed in the following aspects.

As shown in Figure 8, keywords including small vessel disease, connectivity, biomarkers, and lacunar stroke have acquired the recent citation burst, indicating possible future research directions related to VCI neuroimaging. Additionally, cerebral small vessel disease has already been discussed in Part 4.2.1.

This bibliometric analysis still has certain limitations. Firstly, our literature retrieval was limited to the Core Collection of the Web of Science database because it provides complete information beneficial for the data analysis of CiteSpace and VOSviewer. However, this approach might have excluded articles available in other databases. Secondly, we only included literature published in English, potentially leading to selective bias in article screening. Finally, no distinction was made regarding the quality of the included literature when performing this bibliometric analysis, and the lower-quality articles could impact the quality of this study.