Quick Buzz Feed
This bibliometric analysis characterizes global and UK research in generative and foundation-based AI in medical imaging from 2017 to 2025. It identifies 13,452 global publications, with 889 from the UK, ranking the UK fourth in volume but higher in unadjusted citations per paper. The study details publication trends, institutional contributions, collaboration networks, and thematic clusters, highlighting rapid growth and a focus on deep learning, MRI, and diffusion-based applications, while acknowledging limitations of bibliometric indicators for clinical impact.
Artificial intelligence (AI), particularly generative and foundation-based methods like Generative Adversarial Networks (GANs) and diffusion models, has seen rapid growth in medical imaging research. These AI systems offer potential benefits in image acquisition, reconstruction, segmentation, synthesis, triage, and interpretation, aiming to improve diagnostic efficiency and consistency. Foundation models, pre-trained on vast datasets, support generalizable applications across various imaging tasks. The burgeoning literature across clinical, engineering, and conference venues necessitates a systematic understanding of research distribution, key contributors (countries and institutions), collaboration patterns, and dominant thematic areas to inform research strategy and funding. This study employs bibliometric analysis, a quantitative method, to assess research output and citation impact. The United Kingdom is a key focus due to its strong academic medical imaging community, integrated healthcare system, and strategic AI priorities. This analysis aims to comprehensively characterize UK-affiliated research in generative and foundation-based AI in medical imaging from 2017 to 2025, mapping its structure, evolution, and global position.
The methodology employed for this bibliometric analysis began with a detailed search strategy executed on the Scopus database, chosen for its extensive coverage across biomedical, engineering, computer science, and conference literature relevant to AI research in medical imaging. The search encompassed title, abstract, and keyword fields for publications from 2017 to 2025. It combined terms related to generative and foundation-based AI (e.g., "generative AI", "generative adversarial network*", GAN, "diffusion model*", "foundation model*", "large language model*", ChatGPT, "Gemini AI") with medical imaging terms (e.g., "medical imag*", "diagnostic imaging", radiolog*, CT, MRI, ultrasound). A specific filter was applied to identify UK-affiliated publications. Eligibility required at least one AI term and one medical imaging term, with no language or document type restrictions. A validation exercise on a stratified random sample of 400 records indicated a strict search precision of 90.0%. Data cleaning involved exporting bibliographic metadata from Scopus, checking for completeness, and consolidating hierarchical institutional affiliations (e.g., University of Oxford and its medical division). Generic indexing terms were removed before keyword analysis. Descriptive bibliometric indicators such as total publications, total citations, citations per paper, and annual publication growth were calculated. Full counting was applied for country and institution-level metrics. VOSviewer software was utilized for keyword co-occurrence and author co-authorship network analyses, applying specific thresholds for inclusion and using association strength normalization.
The Scopus search yielded 13,452 global records from 2017 to 2025, with 889 (6.6%) having at least one UK affiliation. The search strategy demonstrated a strict precision of 90.0%. Globally, publication output surged, especially after 2022, accumulating 194,650 citations (14.47 citations per paper). China had the highest publication volume, followed by the United States, India, and the United Kingdom. The UK ranked fourth in volume but achieved a higher unadjusted citations-per-paper value of 21.00. UK-affiliated output dramatically increased 18-fold from 17 publications in 2017 to 305 in 2025, consistently accounting for approximately 6-7% of global output from 2019. Journal articles constituted the majority (57.1%) of UK publications, with conference papers also making a substantial contribution (29.4%). Imperial College London was the most productive UK institution with 131 publications, closely followed by King’s College London, University College London, and the University of Oxford. The National Heart and Lung Institute showed the highest citations per paper among the top institutions. Among 3,942 authors, Yang G had the highest publication count (42) and total link strength, while Alexander DC had the highest unadjusted citations per paper (101.27). The co-authorship network revealed five clusters. Over 62% of UK publications reported external funding, with the Engineering and Physical Sciences Research Council, National Natural Science Foundation of China, and UK Research and Innovation being prominent funders. Publications were distributed across 160 sources; Lecture Notes in Computer Science had the largest output, while NeuroImage and IEEE Transactions on Medical Imaging had high citation rates. Keyword analysis identified three thematic clusters: generative and deep learning methodologies, MRI- and diffusion-focused applications, and broader diagnostic imaging workflows. The 20 most highly cited UK publications, collectively receiving 5,389 citations, initially focused on GAN-based methods but more recently included diffusion and foundation-model architectures.
This bibliometric analysis establishes the United Kingdom as a significant contributor to generative and foundation-based AI research in medical imaging between 2017 and 2025. The UK's high unadjusted citation impact, strong involvement from research-intensive institutions, and extensive collaborative networks underscore its prominent role in this rapidly evolving field. This pattern aligns with the inherently multidisciplinary nature of medical imaging AI, which demands a diverse range of expertise including clinical imaging, computer science, engineering, and data governance. The observed mix of journal articles and conference papers in UK output reflects the dual requirements of the field: journal publications facilitate clinical and translational dissemination, while conference proceedings provide a crucial avenue for sharing rapidly developing methodological research. Variations in publication formats across countries likely stem from different disciplinary and dissemination structures rather than solely differences in research quality. Furthermore, the diverse funding profile, encompassing engineering, computational, biomedical, charitable, European, and international sources, highlights the interdisciplinary investment essential for both methodological innovation and translational infrastructure in the UK.
The findings of this study, particularly the sustained increase in publication volume, are consistent with broader bibliometric analyses that have documented the rapid expansion of AI research across medicine and radiology since 2019. Within this growth, generative and deep learning methods have shown particular prominence, as evidenced by frequently occurring keywords like 'deep learning', 'medical imaging', 'diagnostic imaging', and 'GANs', and by the high citation counts of UK-affiliated papers related to GANs, diffusion models, and foundation models. This aligns with previous top-cited analyses identifying deep learning as a key keyword and neuroimaging/oncology as significant clinical subjects. A recent India-focused bibliometric analysis also noted a similar rapid national growth in generative AI research in healthcare imaging, further supporting the relevance of country-level mapping. The keyword and highly cited publication analyses suggest a thematic evolution over time. Earlier highly cited publications primarily addressed GAN-based reconstruction, synthesis, augmentation, anomaly detection, and cross-modality translation, especially in MRI. More recent highly cited works have expanded to include foundation models, general-purpose segmentation approaches, diffusion-based models, and multimodal synthesis reviews. However, it is crucial to interpret this temporal pattern cautiously, as citation rankings are significantly influenced by the age of the publication.
The strong focus on MRI-related applications within the UK research landscape reflects MRI's suitability for tasks such as reconstruction, denoising, synthesis, acceleration, and image-quality enhancement, which address important clinical challenges like acquisition time and workflow efficiency. While bibliometric visibility indicates research attention, it does not directly measure methodological quality, routine clinical adoption, or patient benefits. These require rigorous evaluation through prospective clinical and implementation studies. The potential of deep learning reconstruction to reduce MRI acquisition times is significant, especially in the context of workforce shortages and increasing imaging demand. However, before routine clinical integration, thorough prospective evaluation and clear deployment guidance are essential. Successful implementation faces challenges related to data governance, interoperability, computational infrastructure, procurement, regulatory evaluation, and model monitoring. Generative imaging models, in particular, introduce concerns about artifact generation, image authenticity, and 'deepfakes'. AI software intended for medical use also falls under medical device regulations. Therefore, standardized evaluation and transparent reporting, guided by frameworks like CONSORT-AI and TRIPOD+AI, are crucial. The concentration of UK research in major academic centers and its dense collaboration networks could facilitate multicenter validation and coordinated evaluation efforts. The identified research areas offer potential priorities for future study, but their clinical value must be assessed against patient needs, safety, workflow integration, cost-effectiveness, and equity considerations.
The concentration of research output within highly research-intensive institutions in the UK highlights an established academic foundation but also suggests the potential value in broadening participation across NHS and university centers. Future funding initiatives could strategically support multicenter datasets, cross-institutional validation, and enhanced collaboration among NHS imaging departments, universities, industry, regulators, and patient representatives. This aligns with the UK's AI Opportunities Action Plan, which prioritizes infrastructure development, data access, public sector adoption, and translating AI capabilities into societal benefits. The interdisciplinary funding profile, spanning engineering, computational, and biomedical research, indicates that UK medical imaging AI already benefits from diverse investment. Programs such as UK Research and Innovation's funding for AI in biomedical and health research, emphasizing partnerships and translational impact, provide a relevant model. The Medical Research Council's focus on AI reflects its strategic importance in biomedical research. Critical to translation are secure imaging datasets, trusted research environments, robust computational capacity, interoperability, and clear governance pathways. NHS information-governance guidance underscores lawful data use, transparency, accountability, and appropriate governance, while the Medicines and Healthcare products Regulatory Agency (MHRA) guidance clarifies that medical AI software falls under medical device regulation. Therefore, funding and policy support must balance the development of novel AI models with their rigorous validation, robust governance, continuous monitoring, and comprehensive evaluation across diverse NHS settings.
This study acknowledges several inherent limitations common to bibliometric analyses. Firstly, the analysis was restricted to publications indexed in Scopus, meaning relevant records exclusively found in other databases were not captured. While Scopus offers broad coverage, a single-database approach was chosen for consistency and reproducibility, though it may result in an incomplete capture of the entire literature. Furthermore, due to the retrieval date in February 2026, publication and citation counts for the most recent year, 2025, might be underestimated due to incomplete indexing. Secondly, the findings are dependent on the chosen search strategy. Although designed to be comprehensive for generative and foundation-based AI in medical imaging, the field's rapidly evolving terminology means some narrower or emerging terms might have been missed. Conversely, broad terms may have retrieved some false-positive or indeterminate records, despite a precision check indicating a high accuracy rate. Thirdly, citation-based indicators were unadjusted for factors like publication age, document type, research field, and collaboration patterns. Therefore, 'citations per paper' should be interpreted descriptively, not as a normalized measure of quality or impact, as they can be influenced by journal visibility, self-citation, and field size. Fourthly, the study relied solely on bibliographic metadata and did not evaluate the technical performance, clinical implementation, regulatory status, or ethical aspects of individual AI models. Full counting, used for country and institutional contributions, may overrepresent multi-authored or multi-institutional publications. Lastly, keyword co-occurrence and co-authorship maps are sensitive to chosen thresholds and VOSviewer settings. The lack of systematic synonym merging for keywords could also have impacted their frequencies and cluster composition.
This bibliometric analysis conclusively demonstrates the rapid global expansion of generative and foundation-based AI research within medical imaging between 2017 and 2025. United Kingdom-affiliated research constitutes a substantial and highly cited component of this burgeoning field. The UK secured the fourth rank globally in terms of publication volume and exhibited a comparatively high unadjusted mean citation rate, indicating significant visibility and impact. This research activity is notably concentrated within major academic centers, underpinned by a robust framework of interdisciplinary funding. Keyword analysis and a review of highly cited publications revealed prominent thematic focuses on generative and deep learning methodologies, particularly in the context of MRI reconstruction and enhancement, diffusion-based approaches, and the development and application of foundation-model architectures. These findings collectively provide a transparent and empirically grounded bibliometric reference point for systematically monitoring the ongoing evolution of AI research in medical imaging. However, it is crucial to reiterate that these citation-based indicators, while informative for academic output and collaboration patterns, do not directly quantify or measure clinical implementation, inherent methodological quality, or the ultimate patient-level impact of these advanced AI technologies. Further direct clinical and translational studies are essential to assess these critical aspects.