Exploring Three Decades of Progress: A Comprehensive Visual Review of Intravoxel Incoherent Motion MRI Research (1988–2021)

Background

Microcirculation plays an important role in maintaining homeostasis of the cardiovascular system, delivering oxygen and nutrients, removing waste, and immune system signaling [1–4]. Microvascular dysfunction is associated with numerous diseases [5–7]. Therefore, quantitative assessment of microcirculatory physiology and pathophysiology will help in diagnosis and management of these diseases.

Perfusion imaging, especially magnetic resonance perfusion imaging (MRPI), has become an important means to evaluate microcirculation [8,9]. MRPI has 2 categories: contrast-enhanced (CE) imaging and contrast-free imaging. CE MRI methods, including dynamic susceptibility contrast and dynamic contrast-enhanced imaging, provide quantitative and functional information on the microenvironment of a tissue [10]. However, both techniques require intravenous administration of gadolinium-based contrast agents, which can cause adverse effects such as anaphylactic reaction, nephrogenic systemic fibrosis, and gadolinium deposition in the brain and other tissues [11–13].

The contrast-free MRPI includes arterial spin labeling (ASL) and intravoxel incoherent movement (IVIM). ASL uses magnetically labeled blood water as an endogenous contrast [14]. IVIM was originally defined as “translational movements which within a given voxel and during the measurement time present a distribution of speeds in orientation and/or amplitude” by Le Bihan [15]. Based on diffusion-weighted imaging with a biexponential or more advanced model, perfusion-related effects and diffusion-related effects can be separated by IVIM and the blood-flow properties in capillaries of the interesting tissue can be measured [16].

Because it does not use contrast agents and it provides additional information about microvasculature, IVIM is widely used throughout the body and for various diseases. Especially for oncology, IVIM demonstrated its potential for the differential diagnosis of malignant and benign tumors, or staging of malignant tumors, or the prognosis, or treatment evaluation [17–20].

Despite an increasing number of IVIM articles in the past decade, almost all focused on the technique and clinical applications of IVIM, with little attention paid to the collective knowledge and scientific analysis of this field.

Information visualization is commonly used to reveal the relationships among data and to disclose hidden features. As a branch of informatics, scientometrics uses science-mapping tools to analyze the published literature to build a knowledge structure and explore research hotspots and trends of an interesting field [21]. Compared with other science-mapping tools, CiteSpace software has can perform multivariate statistical analysis, cluster analysis, word frequency analysis, and social network analysis, and focuses on the evolution of trends and research frontiers [22]. Therefore, it has been applied widely in many research fields [23–27].

In this study, CiteSpace was applied to construct a knowledge framework and explore hotspots and emerging trends concerning IVIM in humans from 1988 to 2021.

Material and Methods

DATA COLLECTION:

We searched the Web of Science Core Collection (WoSCC) on 17 August 2021 for relevant articles published 1988–2021, using the retrieval strategy: Topic=(“intravoxel incoherent motion MR imaging” OR “IVIM”) AND English.

The searched articles were refined in document types, and only the data of articles and reviews were downloaded from WoSCC in full-recorded and references, plain-text format. The literature feature clustering included the title, abstract, keyword, source publication, authors’ name, institution, and country, year of publication, and references. Two reviewers independently assessed the downloaded data, with a decision made by the third reviewer if there was disagreement between the first 2 reviewers. Ethics approval and informed consent were not applicable in this study.

ANALYZING METHODS:

The downloaded data were imported into Microsoft (Bellevue, WA, USA) Excel 2016 and CiteSpace V (version 5.8.R1) for subsequent analysis. Excel 2016 was used to draw line graphs to display the number of published articles each year.

CiteSpace V was used to conduct cooperative network analysis for visualizing the collaboration of the authors, institutes and countries, to perform reference co-citation analysis and keywords co-occurrence analysis for studying the knowledge framework and research hotspots, and to perform burst keywords analysis for exploring research frontiers. The parameters of CiteSpace were: Time slicing from January 1988 to August 2021; Term source included title, abstract, author keywords, and keywords plus; Top N=30; and Cluster View-static and show Merged Network in Visualization, which merges the citation network of multiple time periods into a whole network. When the cooperation networks, which show the collaboration among authors, institutions, or countries, were analyzed, “years per slice”, which means the time span for each slice, was set 5 and minimum spanning tree for pruning. For reference co-citation analysis and keywords co-occurrence analysis, “years per slice” was 1 and pathfinder for pruning.

RESEARCH HOTSPOTS AND FRONTIERS:

Research hotspots were high-frequency keywords derived from co-occurrence keywords analysis [28,29] and frontiers explored by keywords burst detection [30]. Hotspot means the key research topic of scientific researchers in a certain period. Frontier means the latest published literature on a research topic, and the terms and phrases contained therein reflect the specific meaning of the research frontier. Research trends are displayed in a keywords time zone map.

Results

ARTICLES SELECTION AND PUBLICATION YEAR:

The total of 1057 articles published between 1988 and 2021 were retrieved by the initial search. After refining with “Article” and “Review”, 991 articles were included. Then, the abstract of each article was carefully reviewed by 2 reviewers, and 67 articles about animal experiments and 3 articles not related to IVIM were excluded. Therefore, 921 articles were included for the further analysis (Figure 1).

According to the publication trend (Figure 2), the number of articles on IVIM was relatively small from 1988 to 2011, with 1–5 articles annually. Since 2012, the number had rapidly grown, from 9 articles in 2012 to 131 in 2017. The yearly number of new publications now appears to have plateaued.

ANALYSIS OF AUTHOR COOPERATION: Network mapping of co-authorship showed that 921 articles were completed by 225 authors (Figure 3); 185 authors published than 5 articles each and the top 10 authors published at least 10 articles each. Christian Federau (18) ranked first, followed by Andreas Boss (17), Moritz C Wurnig (16), Martijn Froeling (14), Bram Stieltjes (13), Queenie Chan (11), Eric E Sigmund (11 articles), Yi Xiang J Wang (10), Changhong Liang (10), and Bachir Taouli (10). Christian Federau, Andreas Boss, and Moritz C Wurnig were in Switzerland; Queenie Chan, Yi Xiang J Wang, and Changhong Liang were in China; Eric E Sigmund and Bachir Taouli were in the USA; Martijn Froeling was in the Netherlands; and Bram Stieltjes was in Germany (Table 1). Their centralities were nearly zero, indicating relatively little cooperation among authors.

ANALYSIS OF INSTITUTION COOPERATION: We found that IVIM studies came from 135 institutions (Figure 4). The top 10 institutions are displayed in Table 2. Philips Healthcare published the most articles (46), followed by Fudan University (30), GE Healthcare (28), Siemens Healthcare (27), and Southern Med University (23). Shanghai Jiao Tong University, New York University, and the Sun Yat Sen University had 21 articles each, whereas Nanjing Medical University and Nanjing University had 18 articles each.

The New York University and the Nanjing Medical University had greater centrality, with 0.41 and 0.20, respectively.

ANALYSIS OF COUNTRY COOPERATION: There were 921 articles published by 43 countries (Figure 5). Among the top 10 countries (Table 2), China had the most articles (392), followed by USA (161), Germany (89), Japan (68), Switzerland (66), France (62), England (52), the Netherlands (44), South Korea (42), and Italy (42). In terms of centrality (more than 0.1), England had the highest centrality (0.42), followed by Switzerland (0.40), France (0.37), Germany (0.36), USA (0.29), Japan (0.16), and the Netherlands (0.11).

CO-CITATION REFERENCES ANALYSIS: There were 15 469 distinct references in co-citation references analysis. We used CiteSpace to create a network map of co-citation references with 654 nodes, 810 links, and main 37 clusters (Figure 6). The modularity Q 0.85 and the weighted mean silhouette 0.96 indicated that the clustering results were reasonable.

In the top 10 clusters (Table 3), “#0 liver” with 68 articles ranked first, followed by “#1diffusion weighting” (46), “#2pancreas” (44), “#3brain” (42), “#4magnetic resonance” (37), “#5experimental and brain ischemia” (32). “#6cardiac”, “#7cirrhosis”, “#8kidney diffusion” and “#9 post-test probability” had 30 articles.

The article published by Andreou A et al [28] was ranked first, with a citation count of 100, followed by Iima M et al [29] (96), Federau C et al [30] (72), and Koh DM et al [31] (69) (Table 4).

ANALYSIS OF CO-OCCURRENCE KEYWORDS: The map of co-occurrence keywords with 281 nodes and 586 links (Figure 7A) demonstrated that the keyword with the highest frequency was “perfusion” (414), followed by “diffusion-weighted imaging” (384), “intravoxel incoherent motion” (338), “apparent diffusion coefficient” (274), and “magnetic resonance imaging” (273), and the one with the highest centrality was “perfusion” (0.40), followed by “diffusion-weighted imaging” (0.34), “apparent diffusion coefficient” (0.26), “magnetic resonance imaging” (0.24), and “tumor” (0.23) (Table 5). The nodes are marked with purple circles representing important keywords.

The co-occurrence keyword clustering graph with an overall Q 0.52 and mean silhouette 0.91 (Figure 7B) showed 13 clusters of keywords. Four were associated with IVIM technique, including “#0IVIM”, “#1mri”, “#3bold”, and “#7epi”, and 9 were related to its application in the body such as “#2preliminary experience”, “#4renal”, “#5glioma”, “#6pca”, “#8renal cancer”, “#9pancreatic cancer”, “#10cerebral blood flow”, “#11kidney diffusion” and “#12pancreas” (Table 6).

As demonstrated by the time zone view (Figure 8), the earliest research direction was “diffusion-weighted imaging” in 1991, followed by “perfusion”, “intravoxel coherent motion”, and “magnetic resonance imaging” in 1992. The second peak keywords appeared from 2008 to 2016, including “parameter”, “b value”, “cancer”, “differentiation”, and “cirrhosis”.

ANALYSIS OF KEYWORD BURST:

The appearance time, burst strength, and time span of the top 17 keywords with the strongest citation bursts varied (Figure 9). From 1988 to 2021, the highest-strength burst keyword was “cirrhosis” (9.68), which began at 2013 and ended at 2015. “Brain perfusion” (3.72) and “accuracy” (4.12) were 2 keywords receiving great attention in recent years.

Discussion

GENERAL INFORMATION:

The trend of IVIM publications from 1988 to 2021 demonstrated 3 periods: the slowly developing period (1988 to 2011) with 48 articles (5%), which focused on the IVIM technique; the rapidly increasing period (2012 to 2017), when 9 articles in 2012 increased to 131 in 2017; and the plateau period (since 2018), with 130 articles per year. This trend was consistent with the development of IVIM techniques. The concept of IVIM was first introduced in 1986 by Le Bihan et al, who used IVIM effect combined with diffusion MRI, but IVIM effect and diffusion MRI was not applied in clinical practice until 1988 [29]. Unfortunately, its application was not popular due to MRI software and hardware limitations. After its coupling with echo planar imaging, the signal at multiple and higher b values could be acquired and IVIM became available [30] and was first used in a series of patients with liver lesions [31].

IVIM studies have been conducted in 43 countries and 135 institutions around the world. Although China and the USA published the most IVIM-related articles, England and Switzerland were central collaborators with other countries. In addition to Philips Healthcare, GE Healthcare, and Siemens Healthcare, 6 of the top 10 institutions were from China, and 1 from the USA. New York University and the Nanjing Medical University played an important role in the cooperation among research institutions. A total of 921 articles were written by 225 authors, but there was relatively little cooperation among authors.

KNOWLEDGE FRAMEWORK:

The co-citation network constitutes the knowledge base [32] and the most highly cited literature was the basic literature in the research field [33]. Thirty-seven clusters were obtained in the co-citation network, of which 15 were associated with IVIM technique such as “diffusion weighting”, “magnetic resonance”, and “dwi pseudo-diffusion”, and 22 were related to clinical applications, mainly on the brain, breasts, heart, liver, pancreas, kidneys, and prostate.

The top 10 references laid the foundation of IVIM research. From the IVIM basic theory and development history of Koh DM [30] and Lima M [28], IVIM imaging proposed by Le Bihan et al [29] was used to quantify the microcapillary perfusion of tissue of interest. To obtain the IVIM parameters, a two-compartment water diffusion model in a voxel was assumed. One was the “microvascular” compartment, where blood water diffusion depends on the velocity of blood and microvascular architecture, and the other was the “nonvascular” compartment, where water diffusion is exquisitely sensitive to the extravascular tissue structure. Given that water diffusion in 2 compartments yields the signal intensity on DWI, biexponential rather than mono-exponential fitting of the data was used to separate the perfusion from diffusion [29].

The b value was a key parameter of DWI and IVIM imaging. Three effects were observed under different b values. When the b value was below 200 s/mm2, the IVIM effect was higher than water true diffusion [34,35]. However, if the b value was above 200–400 s/mm2, IVIM effect was negligible and the Gaussian diffusion effect became stronger. As the b value increased up to 1500 s/mm2 or more, the water displacement distribution deviated from the Gaussian law and the non-Gaussian diffusion effect made the signal reach a “noise floor”, which needed more advanced models to handle this behavior, such as a kurtosis model [36], stretched exponential model [37], and others [38]. To extract 3 IVIM parameters, at least 4 different b values must be used theoretically. However, 4 b values were not possible to evaluate the parameter uncertainties, many studies used 4–10 b values. In addition, Lemke A et al [39] found that b value distributions can substantially minimize the overall measurement error.

Although IVIM effect can be used separately from DWI, it must be kept in mind that IVIM imaging cannot separate blood flow from flowing fluid in a voxel, such as CSF in the brain [40], tubular flow in the kidneys [41], and active transport from glandular secretion (breast ducts, salivary glands, and pancreas) [42].

Seven of the top 10 references reported classic empirical studies on IVIM. Dyvorne et al [43] optimized IVIM imaging by estimating the effects of diffusion gradient polarity and breathing acquisition scheme to detect hepatic fibrosis. Patel J et al [44] reported their preliminary experience in accurate diagnosis of cirrhosis. Woo S et al [45] and Andreou A et al [31] used IVIM parameters to evaluate hepatic carcinoma and metastases, respectively. In 2012, Federau C et al [41] found that IVIM parameters were sensitive to hyperoxygenation-induced vasoconstriction and hypercapnia-induced vasodilatation. Two years later, Federau C et al [30] republished an initial report about IVIM application in stroke, gliomas, metastasis, extra-axial tumor, and status epilepticus. Pang Y et al [46] revealed that IVIM imaging could also diagnose prostate cancer.

RESEARCH HOTSPOTS, FRONTIERS, AND EMERGING TRENDS:

According to the time zone map and keyword clustering, hotspots could be summarized into 3 main hotpots in different time. The hotspot in the first stage, from 1991 to 2000, focused on IVIM theory and technology such as “IVIM”, “mri”, “bold” and “epi”. The hotspot in the second stage from 2001 to 2007 were related to IVIM application in the body such as “blood flow”, “liver”, “model”, and “quantification”. The third stage was from 2008 to 2016, when optimized IVIM imaging had been paid more attention and the frequency of articles using the keywords “parameter” or “b values” increased. Recently, the keywords such as “brain perfusion” and “accuracy” increased dramatically and become a research frontier.

LIMITATIONS:

This study had some limitations. First, all the data were downloaded from the WoS database; therefore, some published studies were not included in this study. Second, all the literature analyzed in the present study were in English and those published in other language were excluded, which might have led to language bias. Finally, this study used co-occurrence keyword analysis to explore the hotspots and frontiers of IVIM research. Weaknesses of co-occurrence keyword analysis [47–49] may have failed to accurately assess some research hotspots, which need further study in the future.

Conclusions

Scientometric analysis of the IVIM literature provide researchers with valuable information about knowledge framework, hotspots, and emerging trends concerning IVIM in humans.