Many meta-analyses have investigated the diagnostic accuracy of common imaging modalities in a broad variety of conditions and body systems. In our study of the 100 top-cited meta-analyses, we found that the key imaging studies relate to cardiac, abdominal or neurological investigations. Furthermore, PET, PET-CT and MRI dominated the imaging modalities utilised. These findings attest to the importance of PET-CT in clinical oncology and research and of MRI in general. The interest in PET and PET-CT, in particular, is likely to be multifactorial and reflective of the increased use of PET for cancer staging , technological advancements in PET imaging and the development of new radiotracers as well as increasing interest in the clinical utility of hybrid and molecular imaging techniques [13, 14].
While it was unsurprising that studies focusing on cancer diagnosis and staging had the highest citation counts, it was interesting to note a lack of studies associated with lung cancer imaging. This could be reflective of peaking incidence rates of lung cancer in some Western countries due to tobacco control measures  and relatively less interest in new diagnostic tests compared with other malignancies (although we note the increasing interest in lung cancer screening using technologies such as low-dose CT ).
The most commonly cited studies had years of publication that appear to cluster around the period from 2008 to 2014. This may point towards an increase in interest towards diagnostic-accuracy-based meta-analyses in the field of radiology in these years or increasing confidence in the statistical techniques used in such studies (e.g. hierarchical models, which are more complex than methods used in meta-analyses for therapeutic interventions and increasingly used and recommended ). More recent studies may have lower citation counts as they have had less time to be of influence.
We also found that less than half of the meta-analyses compared one imaging modality against another. While such comparative analyses may be more useful in assessing new imaging tests against established modalities and diagnostic pathways, this suggests researchers may be more interested in studies of single modalities. This may reflect the fact that meta-analyses of diagnostic test accuracy, in general, often focus only on a single test and those which do address comparative test accuracy may be poorly designed or biased and difficult to interpret 
To our knowledge, this is the first bibliometric analysis of meta-analyses concerning diagnostic accuracy in radiology. In contrast, bibliometric studies of the general radiology literature are plentiful, for example [19,20,21]. In these studies, citation counts for the 100 top-cited articles ranged from 624 to 6447 , 371 to 6931  and 422 to 7506 . Even ignoring differences in publication dates, given that the top-cited meta-analysis was associated with 394 citations , meta-analyses of diagnostic accuracy are unlikely to feature substantially in these, or future bibliometric studies of the broader radiology literature. Our study may therefore provide a potential resource for identifying key diagnostic accuracy information related to imaging modalities used in the field of radiology which may be of particular interest to researchers, academics and practising radiologists with an interest in evidence-based diagnosis.
Although there are some differences between databases that report citation counts such as Web of Science, Scopus and Google Scholar , we elected to use Scopus due to the availability of an Application Programming Interface (API). This allowed efficient and complete extraction of author names, affiliations, publication details and citation metrics. One particular advantage of the Scopus database is the Affiliation Identifier that assigns each institution a unique number, thereby enabling aggregation of multiple affiliations from the same institution.
A key focus of bibliometric analysis is citation counts. Importantly, citation counts are influenced by a multitude of factors . Some of these factors such as structured abstracts  and study design  may reflect the quality of reporting and strength of evidence. In contrast, other factors such as open access [26,27,28], and title length  may instead be related to visibility and accessibility. The association between citation counts and quality therefore needs to be considered carefully, especially given the risk of citation bias [6, 7], the preferential citing of statistically significant results that may lead to inflated expectations of efficacy. Certain studies may also be preferentially cited based on the reputations of the journals they are published in, even if they may have comparable quality to those published in lesser known journals. As this was a bibliometric study, we did not review the quality and reporting of meta-analyses and these considerations have been addressed elsewhere [2, 29,30,31]. The recently published guideline for reporting meta-analyses of diagnostic test accuracy, the PRISMA-DTA statement , provides a 27-item checklist for reporting such studies. Compliance with this reporting framework may be the subject of future investigation.
A related issue is that while citation counts, in this instance, may be a useful metric for gauging the extent to which the research community is interested in particular imaging modalities, they do not necessarily reflect clinical outcomes of these modalities or the feasibility of adopting new imaging technologies in clinical practice. This is especially relevant to meta-analyses of diagnostic accuracy which invariably focus on technical measures of diagnostic performance and not on parameters that may be of more clinical interest such as cost-effectiveness, ease of access or patient and clinician acceptability.
Our study had some limitations. First, the Scopus API was the primary tool used for extracting citation counts, authorship, and affiliations. We noted that the affiliations recorded in the retrieved document and affiliation profile were an aggregate of those listed in the published articles. While such an approach prevented the discrimination between first and corresponding author affiliations, it instead afforded the opportunity to capture all distinct institutional affiliations. Thus, in our analysis we did not distinguish between the affiliations of different authors. Second, our search was not limited to radiology-specific journals in the category “Radiology, Nuclear Medicine and Imaging” by Journal Citation Reports (Clarivate Analytics, 2018) . Instead, we included every journal in the category. Notably, despite conducting a broad search, the meta-analyses retrieved were predominantly published in radiology journals, whereas only several papers were published in cardiovascular imaging or nuclear medicine journals. However, by restricting the search to a single category, relevant meta-analyses published in other general medicine or specialty journals may not have been included. Third, differences in citations may exist between different search databases . We mitigate this by including an extensive list of meta-analyses, that is of the 100 top-cited papers. Fourth, we restricted our search to meta-analyses due to the fact that these provide concise and readily interpretable summary information of a quantitative nature. However, both systematic reviews and meta-analyses are widely considered to sit at the top of the hierarchy of evidence. While many meta-analyses incorporate a systematic review in order to identify all relevant primary studies, not all systematic reviews incorporate meta-analysis (due to lack of primary studies or significant underlying heterogeneity). Therefore, it is possible that important evidence syntheses may not have been identified by our current work.
There are newer means of assessment of the influence of scientific publications such as Altmetrics which provide information on citations in the broader online and media community and may be complementary to traditional, citation-based metrics. While it is presumed that imaging-based diagnostic accuracy meta-analyses might receive less attention in these domains due to their technical nature, some studies of influential imaging modalities may receive wide attention in the media and this impact has not been evaluated here.
In conclusion, our bibliometric study provides a collation of the most influential meta-analyses of diagnostic accuracy in radiology as measured by citation counts. The 100 top-cited meta-analyses encompass a broad range of imaging modalities and body regions. However, we emphasise that citation count and quality should not be confounded, and that individual papers be considered on their own merits.