Global research trends in COVID-19 with MRI and PET/CT: a scoping review with bibliometric and network analyses

Objective To identify and evaluate the indexed studies that allow us to understand the implications of imaging studies in MRI and PET/CT related to COVID-19 research. Methods Scoping review. Articles in PubMed, Scopus, and Web of Science (WoS) were scanned from 2019 to 2021 with COVID-19, MRI, and PET-CT as keywords. EndNote software and manual checking removed the duplicated references. Our assessment includes citation, bibliometric, keyword network, and statistical analyses using descriptive statistics and correlations. Highlighted variables were publication year, country, journals, and authorship. Results Only 326 papers were included. The most cited article reached 669 cites; this number represented 21.71% of 3081 citations. The top-15 cited authors received 1787 citations, which represented 58% of the total cites. These authors had affiliations from ten countries (Belgium, China, France, Italy, Japan, Spain, Sweden, Turkey, United Kingdom (UK), and the USA). The top-30 journals were cited 2762 times, representing 89.65% of the total cites. Only five journals were cited more than 100 times; Int J Infect Dis had the most significant number of citations (674). Some of the unexpected keywords were encephalitis, stroke, microbleeds, myocarditis. Conclusion COVID-19 pandemic is still spreading worldwide, and the knowledge about its different facets continues advancing. MRI and PET/CT are being used in more than 50% of the selected studies; research trends span seven categories, no only the diagnostic but others like socio-economic impact and pathogenesis Developed countries had an advantage by having hospitals with more resources, including MRI and PET/CT facilities in the same institution to supplement basic assessment in patients with COVID-19. Supplementary Information The online version contains supplementary material available at 10.1007/s40336-021-00460-x.


Assessing the influence of publications in medical research
An accurate assessment to determine the influence of medical papers is fundamental for improving physicians' ability to identify essential articles in a specific scientific research field. This activity also guides clinical practice decisions; as the number of articles that researchers need to be familiar with increases, so does the importance of selective searching and generalization [1].

Coronavirus disease 2019 (COVID-19)
Two well-known b-coronaviruses caused severe disease with considerably high mortality rates in recent memory, the severe acute respiratory syndrome coronavirus 1 (SARS-CoV-1) and Middle East respiratory syndrome (MERS) epidemics [2,3]. The novel SARS-CoV-2 b-coronavirus driving the current coronavirus disease 2019  pandemic has depicted transmission rates far above those seen during prior b-coronavirus epidemics [4]. COVID-19 emerged in December 2019 and rapidly developed into a global outbreak [5].
COVID-19 patients may be asymptomatic or present no specific signs and symptoms like fever, dyspnea, and cough; this lack of specif clinical features can increase the risk of spread of the epidemic [6]. It is accepted that the clinical manifestations of COVID-19 show a significant overlap in presentation with other respiratory tract infections, such as the influenza virus, which makes its diagnosis based on clinical grounds challenging [7], especially during the year when the flu is epidemic [8]. SARS-CoV-2 infection cannot be entirely excluded, even if reversetranscription polymerase chain reaction (RT-PCR) results are negative [9]; also, a high false-negative rate has been reported, which can cause a miss or delay in the effective diagnosis [10]. Therefore, medical imaging is a complementary examination in the diagnosis and management of COVID-19 [11].

MRI and COVID-19
Although radiologic imaging's utility in screening and diagnosing suspected COVID-19 cases has been hotly debated since the start of the pandemic [7], chest imaging has played an essential role in both the case definition and clinical classification of COVID-19 [11]. In day-to-day practice, radiologists stratify cases of COVID-19 based on radiologic imaging to identify those patients most at risk for imminent clinical decompensation [12,13].
At least three studies have compared pulmonary magnetic resonance imaging (MRI) features of COVID-19 pneumonia against conventional chest computed tomography (CT). It is possible the visualization of groundglass opacities (GGOs), consolidation, reticulation, and a reverse halo sign on multiple MRI sequence [14], with no significant differences in MRI detection of GGOs or consolidative lesions versus conventional CT [15]. Ultrashort echo time MRI (UTE-MRI) vs. conventional CT has demonstrated high concordance in lesion detection [16]. Current guidelines from the American College of Radiology recommend limiting the use of MRI in confirmed or suspected SARS-CoV-2-positive patients [16,17]. MRI may provide a viable alternative for imaging high-risk patient groups (pregnant patients and children, in whom exposure to ionizing radiation should be avoided) [14-16, 18, 19].

PET and COVID-19
Recent studies have evinced a growing role in using positron emission tomography (PET) to evaluate and characterise infectious and inflammatory pulmonary conditions [20][21][22]. A recent study that assessed the presence and pattern of incidental interstitial pneumonia due to COVID-19 on fluorine-18fluorodeoxyglucose positron emission tomography/computed tomography (PET/CT) in asymptomatic oncological patients, found a rate of 7.1% during the COVID period compared with 5.35% pre-COVID period; the detection rate was 59% with a maximum standardized uptake (SUVmax) of 4.1 [6].
It is known that acute lung injury in viral pneumonia is driven by chemokine recruitment of neutrophils, monocytes, and effector T cells [21,23]. Neutrophil activity manifests as 18 F-FDG-avid foci on PET imaging [7]. Viral replication occurs in the absence of clinical manifestations of the disease [24]. Therefore PET/CT has been proposed as an alternative modality to detect early, subclinical infections, albeit with an associated radiation cost [21,23]. COVID-19 pneumonia has shown parenchymal 18 F-FDG uptake in regions corresponding to GGOs and consolidative opacities, with maximum standardized uptake (SUVmax) values ranging from 4.6 to 12.2 [25]. This radiotracer uptake has also been noted in the hilar and paratracheal lymph nodes and the bone marrow [26]. A systematic review reported a mean SUVmax of 4.9 ± 2.3 in COVID-19-associated pulmonary lesions [27]. Two radiotracers used in the setting of prostate cancer, the 68 Ga-labelled prostate-specific membrane antigen ( 68 Ga-PSMA) and 18 F-labelled choline ( 18 F-choline), have identified hypermetabolic lesions corresponding to subpleural GGOs [28]. A greater 18 F-FDG uptake may correlate with higher erythrocyte sedimentation rates and a longer time to heal [21,25,29]. 18 F-FDG PET/CT may also help assess for changes in other organ systems, including the heart, kidneys, and gastrointestinal tract [29], and the presence of concomitant infections in situations of diagnostic uncertainty [30].

Purpose of the study
We found more than 83,000 articles, including the phrase COVID-19, in the title of articles indexed in the National Library of Medicine (PubMed) of USA until April 7th of 2021. A systematic review of the research trends in MRI and PET in COVID- 19 has not yet been described. We aimed to identify and evaluate the indexed studies that allow us to understand the implications of research imaging in MRI and PET/CT articles related to COVID-19 research.

Scoping review
A scoping review describes existing literature and other sources of information and commonly includes findings from various study designs and methods [31]. Its purpose is to provide an overview of the available research evidence without producing a summary answer to a discrete research question [32]. For developing this study, we followed the Arksey and O'Malley approach framework, which summarises five stages for reporting a scoping review [32]: (1) identify the research question; (2) identify relevant studies; (3) study selection; (4) chart the data; and (5) collate, summarise, and report the results. Levac et al. [33] and the Joanna Briggs Institute [34] updated the methodological framework. This study also was abided by the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) Checklist [35] and recent methodology for reporting systematic quantitative reviews [36]. This data-driven analysis did not require ethical approval, all data are in the public domain and collated data are included in this paper.

Identifying the research question
Considering that radiological tools such as chest X-rays and CT in detecting COVID-19-related interstitial pneumonia have already been accepted as imaging criteria for the final diagnosis [37][38][39]; we wanted to know the helpfulness of other imaging methods that were avaible in tertiary care hospital during the COVID-19 pandemia. Then, our question was: What is known from the existing research in the field of "imaging in COVID-19 using MRI and PET/CT"?

Identifying databases and relevant studies
Articles were retrieved from PubMed, Scopus, and Web of Science (WoS) with the period restricting from 1st January 2019 to April 9th 2021 (the date when we performed the search on the selected databases), using the search terms "COVID-19", "MRI", and "PET/CT".
We included independent search algorithms considering the selected databases' specific search terms and commands (Supplementary file, Table A).

Study selection
Two reviewers selected the articles with predefined inclusion and exclusion criteria (Supplementary Table A) independently (N.R.S. and R.G.V.A.); any discrepancy was resolved by discussion. Initial filtering methods included the period (2019-2021), and language limited to English, subsequent filtering through title and abstract to exclude the irrelevant articles, the final screening was done by reading the fulltext version of each article. After running search strategies specific for each database, we removed the retrieved articles' duplications using the EndNote ® software (version X9.3.3, © 2020 Clarivate Analytics, Boston, MA, USA). By following the recent guidelines for Scoping Reviews [35], the study's flowchart is presented in Fig. 1.

Articles demographics, citation and bibliometric analyses
Data regarding journals, publication year, country and author were collected. Qualified articles were ranked in descending order by the citation number.

Journals' bibliometric analyses
Six bibliometrics corresponding to the journals of selected publications were analyzed to evaluate journals' impacts: Impact factor (IF), Eigenfactor score (ES), CiteScore, Source Normalized Impact per Paper (SNIP), SCImago Journal Rank (SJR) and Quartile (from Q1 to Q4). Journals with a higher score of these indicators are generally considered more prestigious than those with lower scores [40].

Global research trends in COVID-19
Three variables were recorded to evaluate this field: type of article, research direction, and imaging tool. The type of article included seven categories: (1) prospective study; (2) retrospective study; (3) randomized clinical trial (RCT); (4) case report or case series; (5) basic science research; (6) consensus and guidelines, and (7) case and control studies.

Network analysis of keyword cooccurrences
Network analysis is used to reveal the scientific structure, the degree of subject correlation, literature retrieval, and others, starting from the topic correlation reflected in the citations. We performed a network analysis of the keyword correlation of selected publications. This keywords analysis can identify a research topic by closely related publications in citation relations [1]. Network analysis allows the detection of current research hotspots in the field of interest.
VOSviewer (version 1.6.16, Leiden University, The Netherlands) was applied to perform network analyses. We selected "all keywords" as the unit of analysis and set the minimum number of occurrences of a keyword to 5. VOSviewer allowed us to build network analysis of bibliometric relations using large-scale data. We used the density visualization mode. The yellow nodes indicate increased weights of the neighbouring items; the node changes toward blue as the neighbouring items' weights decrease. The size of the node increases with the number of items in the neighbourhood [41].

Software used for database creation, statistical analysis and data visualization
We used Microsoft Excel v16.33 (Microsoft Corporation, Redmon, WA, USA) to manage the database created with EndNote. The IBM ® SPSS ® Statistics software (version 26.0.0.1 IBM Corporation; Armonk, NY, USA) and JMP ® Pro software (version 14.3, SAS Institute Inc., Cary, NC, USA) helped us designing frequency tables, bar graphs, parallel coordinates plot and Spearman correlations analysis to compare correlations between bibliometrics of the journals of selected articles. The methodology for using a correlation matrix in bibliometrics has been previously described [42]. Statistical significance was indicated by p < 0.05 (two-tailed).

Selected articles
Six hundred and twenty articles were retrieved from Pub-Med, Scopus, and WoS (after filters in each platform). The three databases yielded 419, 367, and 280 papers, respectively. The EndNote software removed 210 duplications; 84 publications were excluded after screening the title, abstract and reviewing the full text according to the inclusion or exclusion criterion. Only 326 papers meeting all identified requirements were included. Figure 1 shows the selected papers based on the PRISMA-ScR guidelines. The complete list of selected articles is included as Table  B of a supplementary online-only file.

Total citations analysis
The Kolmogorov-Smirnov and Shapiro-Wilk tests for normality showed a non-normal distribution of the total citations. For that reason, we did not use the mean and standard deviation but the median, quartiles and interquartile range (IQR) to describe central tendency values for this variable. The median for the total cites was 2 with a 75-percentile value of 6;

Date of publication analysis
PET and MRI publications in the context of COVID-19 appeared as early as February 2020. The average number of publications per month was 22. Since August 2020, the number of publication had been above the mean, but with an apparent decremental trend since January 2021. Figure 2 shows a bar graph with the number of publications per month.

Country of publication analysis
Forty-seven countries from all continents published the 326 selected articles, the top-five countries with publications were the USA (82), Italy (31), France (28), United Kingdom (28), and Spain (24); the sum of these articles represented 59.3% of the 100% (326) selected articles. A frequency table of countries is included in Table E in the supplementary online-only file. Figure 3 shows a global map of the published articles around the world.

Journals' analyses of total citations, number of articles and Average no. of citations per paper
We recorded 166 journals from articles that fulfil the inclusion criteria; from those, only 108 journals had at least one citation. The mean number of citations for all journals was 28.53, and the range was from 1 to 674. The top-30 journals were cited 2,762 times, representing 89.65% of the total cites; the range of citations for those journals was from 14 to 674.
Only five journals were cited more than 100 times; Int J Infect Dis had the most significant number of citations (674), followed by Radiology (469), Eur J Nucl Med Mol Imaging (216), Brain (207) and Acta Neurochir (Wien) (110). When we sorted the journals by descending the number of average citations, IEEE Rev Biomed Eng entered the top five journals. If we sorted the journals by descending the number of articles, the list of the top five journals changed to Eur J Nucl Med Mol Imaging (23), Clin Nucl Med (22), Radiology (9), BMJ Case Rep (9), and Acad Radiol (7).
Considering the 108 journals that received at least one citation, there was a moderate strength of Spearman correlation coefficient between the number of citations of each journal and their number of published articles (Rs = 0.525, p < 0.001). Table 2 shows the top-30 cited journals in COVID-19, MRI and PET/CT research.

Journals' analyses of bibliometrics in the top-30 selected papers
Because the six selected bibliometrics depicted non-normal distribution, we reported the median, quartiles and IQR. Most of the journals belonged to Q1; 50% of the top-30 cited articles have an IF above 3.38. The median for the CiteScore was 4.6 and for the Eigenfactor 0.010770. Table 3 shows the descriptive statistics of the bibliometrics from the top-30 cited journals.
The impact factor had a range between one and eleven; the Eigenfactor score for most of the journals remained below 0.0200. The CiteScore seemed to have grouped values within two ranges 0 to five and 10 to 15. The SNIP had a widespread between 0 and 3. Values of SJR were primarily concentrated between 0.5 and 3.5. Most of the All bibliometrics depicted a positive, mild, to moderate correlation with the total citations. However, except for the impact factor, the correlation between bibliometrics and No. of published articles in each journal depicted a very low or even negative correlation. Figure B, in the supplementrary file, shows scatter plots of the bibliometrics vs total citations correlations and bibliometrics vs No. of published articles.
The strongest, positive correlations (between 70 and 90%) were observed between the Quartile, SJR and SNIP, with the other metrics. We only found four negative correlations between bibliometrics that were not significant (CiteScore ⇔ No. of articles, SNIP ⇔ No. of articles, SJR ⇔ No. of articles, and Quartile ⇔ No. of articles). Table G presenting all the Spearman correlation coefficients and p-values is included in the supplementary online-only file.

Global research trends in COVID-19
We found that 54% of the studies corresponded to the case report and case series category, 26.8% were retrospective and only 13.1% prospective studies. About the imaging method used, 41.8% reported only MRI, 29.9% used only PET/CT, AND 25.6% of the studies used both imaging techniques. Figure 4 shows bar graphs of the type of study and research directions in COVID-19 using MRI and PET/CT.

Network analysis of the keywords
We performed a separate network analysis for PubMed, Scopus, and WoS to compare each database's intrinsic variations. Each word was counted once, no matter how many times it appeared in the same article. For the PubMed database, 73 words met the threshold (occurred five times at We use the network visualization and density visualization modules of the VOSviewer software. In the density visualization modules, we selected the item density option and rainbow colour pattern. Figures 5(A, C, E) show the score of each node by its colour. Links indicate keyword relevance, and the number of keyword occurrences increases the node size. Nodes are proportional in size and importance.
Graphs from the density visualization module in Figs. 5(B, D, F) show each node's score by its colour. Different colours represent the time when the keyword appears. Red nodes indicate increased weights, and increased size indicates more significant numbers of neighbouring items. Light blue nodes indicate decreased weights and numbers of the neighbouring items. In Fig. 5, the keyword networks are presented related to their original database (A and B, networks from PubMed; C and D, networks from Scopus; E and F networks from WoS).

Goals and selected articles
In this study, we achieved the aims posed in the introduction; we identified target articles in the three databases analyzed. Then we created a unique database, removed duplicates, and then analysed the selected articles at different dimensions: bibliometrics, demographics, authorship and research trends. From the original 620 articles retrieved, only 52.6% were selected for the analysis.

Citation and publication analyses, authorship and mapping of publications
Citation analysis allowed us to understand the COVID-19 historical perspective and progress of related publications. We can identify seminal papers in the early stage that have impacted the field and act as a surrogate to measure the productivity and influence of a scientist or institution [59]. Our finding of a median of 2 citations per article is similar to the average citation of 2.76 per document reported for studies about COVID during the first 4 months of 2020 in the Scopus database [60].
Although the publication dates spanned only 15 months, that was enough for some articles to be cited more than 100 times. Almost 70% of the articles were cited in that short period, although 30% of the articles did not cite at all.
The most cited article was published in a journal (Int J Infect Dis) with an impact factor of 3.202. Authors do not need to publish in a journal with high IF to received citations; also, high IF does not warrant citations. However, it is the content itself of the manuscript that the readers will evaluate as valuable or not.
Neither the IF nor the total number of citations is, per se, the metric of the overall influence of a journal. All journals have diverse citations, and even the best publications contain some papers that are never cited [61]. Citations are not equally distributed, with fewer than 20% of the articles accounting for more than 50% of the total number of citations [62,63].
There were no previous studies of MRI and PET/CT in COVID-19 to compare our findings. A recent study determined that the scientific quality of COVID-19 publications in three of the highest ranked scientific medical journals (NEJM, JAMA, The Lancet) is below the quality average of these journals [64].
Future studies will address if the downtrend in the number of publications detected for the first semester of 2021 persists in the middle term. Almost 60% of the published articles were produced in some countries more affected by COVID-19 (USA, Italy, France, United Kingdom, and Spain).
It is now clear for researchers that the number of people infected is much greater than the number of reported cases. With more publications on our chosen topic, affected people outnumber detected people by one to two orders of magnitude in Spain, the UK, Italy [65]. The three most cited authors had nationalities from Japan, UK and China, which coincides with the most influential countries with the highest Citation Impact related to COVID-19 and Mental Health [66]. The preponderance of articles from developed countries reflects the ample financial resources available to support research by the scientific community.

Journals' bibliometric analyses
Selected journals had a median IF of 2.488. The average citation per journal was 28.53; it was above the number of 23.4 citations per article, reported by a study of 2530 publications about scientific impact in the first months of COVID-19 [67]. These numbers are pretty below a previous report of the top-50 cited COVID-19 related articles in which 18% of articles received more than 2000 citations; also, the general topic of COVID-19 concentrated half of the top-50 cited publications in only three journals (The Lancet, NEJM, and JAMA) [68]. The most cited journal in our study (Int J Infect Dis) received 674 citations; this number is below the 1192 cites received by the Am J Emergency Medicine in a study that reported top emergency medicine journals in terms of COVID-19 publications in 2020 [69].
Strong, positive correlations had been previously reported between total cites and the six selected bibliometrics for the Radiology, nuclear medicine and molecular imaging journals [42]. A correlational study between the number of articles published per journal and their bibliometrics is missing in the literature.
This trend has also evolved to assess the impact of specific medical topics like achalasia [1], diverse techniques like robotic surgery [78], and of course, the early impact of COVID-19 [64,68]. However, to the best of our knowledge, there was no previous publication of MRI and PET/ CT research trends in COVID patients.
It was not a surprise that most of the studies were directed to diagnosis, clinical characteristics and complications. Considering MRI and PET/CT and imaging techniques belonging to the speciality of Radiology and medical imaging, there is a dominant role of MRI for reporting these studies; however, ¼ of the studies is using both imaging techniques, which is a significant achievement for those hospitals than can afford both imaging techniques in the same patient. We are still in an early stage of the pandemic to visualize future directions from a global perspective. However, independent studies started publishing about risk stratification and prognosis based on MRI findings [79].
We agree with previous reviews that, PET/CT cannot substitute or integrate high-resolution CT to diagnose suspicious COVID-19 or for disease monitoring, also, we learned that up to 48% of Nuclear Medicine department [80], reported that PET/CT incidentally detected suspicious COVID-19 lesions in patients undergoing standard oncological and non-oncological indications [81]. Also, COVID-19 had an impact on Nuclear Medicine departments' daily activities, quantitatively reducing diagnostic and therapeutic procedures up to 51% [80], and by adapting the security protocols to international guidelines [82]. We believe that medical research will include in the short term the analysis of economic networks into the complexity of COVID-19 emergency [83].

Keyword network analysis
We observed that each database identified unique keywords that did not appear in other databases and that those keywords were linked to specific articles related to MRI or PET/ CT. In the Total link strength table, the words that caught our attention were PubMed [incidental findings [84], adolescent [85]]; Scopus [case report [86]]; and WoS [encephalitis [87], stroke, [88] microbleeds [89], myocarditis [90]].

Limitations of the study
We only included articles published in English from the WoS, Scopus and PubMed databases; with this approach, we may have lost sight of influential articles published in other languages and included in other databases. This strategy may imply some degree of selection bias if we could not identify all available data on the topic; however, we intentionally excluded reviews and grey literature. Readers should be aware that the critical appraisal of the risk of bias in scoping reviews is not considered mandatory [96]. Systematic reviews commonly select the information sources by requiring specific study types, such as RCT, and imposing quality standards, such as adequate allocation concealment, and emphasise synthesizing data to address a specific research question [96]. On the contrary, a scoping review describes existing literature and commonly includes findings from various study designs and methods; in our case, we included case reports, case series and guidelines [31]. The synthesis of findings from individual studies and the generation of 'summary' findings is not mandatory in a scoping review [97]. Since the scope of our study was relatively wide, it was not able to perform a systematic quantitative review as Guzman-Ortiz et al. conducted [36].
To avoid the bias of time on citation number, we include all articles that meet our inclusion criteria, even those with zero citations; we wanted to prevent the exclusion of articles published in the last months that, because they still did not receive citations, they were not going to be in the list of selected papers. We acknowledge that our per-month analysis of publication and total citations for papers published in the last 15 months is a snapshot of publications that, over time, many of them may accrue enough citations and move into the top 100 articles. Citation of scientific papers usually begins 1-2 years after publication generally reaches a maximum after 3-10 years and then tapers off to a lower rate [59]. It may take two decades to more accurately assess the true impact of a publication [98].

Conclusions
COVID-19 pandemic is still spreading worldwide, and the knowledge about its different facets continues advancing. MRI or PET/CT is used in more than 50% of the selected studies. We identified research trends spanning seven categories, no only the diagnostic but others are emerging like socio-economic impact and pathogenesis. There are already some articles with more than 100 citations that can be considered seminals in this field. Developed countries had an advantage by having a hospital with more resources and MRI and PET/CT resources in the same institution to supplement basic assessment in patients with COVID-19.