Artificial intelligence in pediatrics: a bibliometric analysis of global output, networks, and frontiers [2016–2025]

Introduction

The rapid development of artificial intelligence (AI) technologies, including machine learning and deep learning, has significantly impacted the way that diseases are predicted, diagnosed, and treated in pediatric populations. The evidence demonstrates AI’s potential through several concrete applications. Ghunaim et al. (1) show that AI tools like DeepVariant have improved genetic anomaly detection precision, while Ganatra et al. (2) highlight AI’s success in early disease detection, such as identifying sepsis and supporting diagnostic imaging. Kerth et al. (3) systematically reviewed 31 studies and found AI applications diverse, including disease classification and outcome prediction, though most current research remains small-scale. Carroll et al. (4) caution that while AI shows immense promise, significant challenges remain in ensuring data quality, patient privacy, and addressing potential algorithmic biases. The technology is transitioning from theoretical potential to practical implementation, but it requires continued rigorous validation and ethical consideration. Despite the growing body of literature on this topic, a systematic analysis of the research landscape is necessary to understand the current status and future prospects. This study seeks to fill this gap by employing bibliometric methods and the CiteSpace software to analyze the literature retrieved from the Web of Science Core Collection (WoSCC) database from 2016 to 2025.

The bibliometric analysis of the AI application in pediatric diseases is crucial for several reasons. Firstly, it helps to identify the key research areas that have received substantial attention from the scientific community (5). Secondly, it highlights the most influential authors and institutions that are driving the field forward (6). Thirdly, it provides insights into the thematic clusters and research trends that are emerging within the domain (7). Lastly, it aids in predicting future directions for research and identifying potential gaps that need to be addressed (8). Existing bibliometric analyses of pediatric AI have focused on single subspecialties with extended historical coverage: Shu et al. examined pediatric surgery from 1995 to 2023 (5), and Velasco Cardona et al. [2025] analyzed bacteriology from 2007 to 2024 (7). The present study adopts a focused contemporary approach [2016–2025] to capture the deep learning era, specifically analyzing publications across all pediatric disease domains—enabling identification of cross-domain knowledge clusters and field-level dynamics invisible to subspecialty-restricted approaches. This study will contribute to the existing literature by offering a detailed overview of the research on AI applications in pediatric diseases. The findings will be valuable for researchers, policymakers, and healthcare professionals who are interested in the role of AI in improving pediatric healthcare outcomes. The analysis will also serve as a foundation for future studies that aim to explore specific aspects of AI applications in pediatrics in greater depth. We present this article in accordance with the BIBLIO reporting checklist (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0116/rc).

Methods

Data source

The data for this study were sourced from the WoSCC database, selected for its rigorous indexing standards and comprehensive coverage of high-impact peer-reviewed literature. While databases such as Scopus, PubMed, Embase, and IEEE Xplore offer broader coverage, they also include sources of more variable quality; we therefore prioritized established peer-reviewed publications to ensure methodological rigor and reproducibility. We analyzed research articles published over the past 10 years that focused on the application of AI technology in the prevention, diagnosis, and treatment of pediatric diseases. The search spanned from January 1, 2016, to the present, with the search conducted on November 30, 2025. The search query was TS = ((“artificial intelligence” OR “deep learning” OR “machine learning”) AND (“pediatric*” OR “paediatric*” OR “neonat*” OR “infant*” OR “adolescent*”) AND (“disease*” OR “disorder*” OR “illness*” OR “syndrome*”) AND (“prevent*” OR “diagnosis” OR “treatment*” OR “prognos*”)). The search terms were developed through a systematic review of standard terminology in the field, cross-referencing with published bibliometric studies on AI in healthcare, and iterative refinement to balance sensitivity and specificity. The final search string combined established AI terminology with pediatric population terms and disease-related keywords to capture clinically relevant applications. We verified our retrieval trends by comparison with published studies in the field, which showed comparable publication patterns and thematic distributions. A total of 1,927 relevant articles were retrieved. These articles were exported in Plain Text File format and then manually reviewed by three researchers who examined the titles, abstracts, and keywords to screen the literature. The inclusion criteria covered experimental studies, observational studies, and reviews. The exclusion criteria included editorials, conference papers, popular science articles, research proposals, press releases, duplicate publications, and articles unrelated to the topic. Specifically, 25 non-research articles were excluded first, followed by two non-English publications and seven off-topic articles, yielding in the final inclusion of 1,893 articles, as illustrated in Figure 1.

Figure 1 Flow chart of document screening. This flowchart illustrates the screening process for publications retrieved from the WoSCC [2016–2025]. WoSCC, Web of Science Core Collection.

Statistical analysis

The annual distribution of publications, core authors, countries, institutions, journals, and keywords was statistically analyzed using Excel 2016 software. The retrieved data were subjected to visualization analysis using CiteSpace 6.4.R1 software. CiteSpace is a bibliographic data visualization software designed and developed by Professor Chaomei Chen’s team at Drexel University in the USA, and has been widely applied in bibliometric statistical analysis.

The literature from the WoSCC database was imported in “plain text file” format and stored in the input folder. After deduplication, the data were exported and stored in the output folder, and then the deduplicated txt files in the output folder were integrated into the data folder. The data source type was set to WoSCC, and the analysis types were selected as author, institution, country, keyword, and reference, respectively. The frequency of term occurrence was then set to generate the required maps. CiteSpace parameter configuration: time slicing 2016–2025 with 1-year intervals; selection criteria set to top 50 most cited items per slice; pruning method using Pathfinder network scaling. Keyword clustering was performed using the log-likelihood ratio (LLR) algorithm. The modularity Q value was 0.4214, indicating significant clustering structure, and the mean silhouette score was 0.6445, indicating good cluster homogeneity. Burst detection was conducted using Kleinberg’s algorithm with γ=1.0 and a minimum duration of 2 years, identifying 28 burst keywords.

Results

General overview

As shown in Figure 2, the number of publications on the application of AI in pediatric disease-related research has been increasing year by year in the past 10 years (R²=0.9301), with the highest number of publications reaching 526 in 2025. Table 1 presents the top 10 most cited articles among the 1,893 retrieved publications, ranked by their citation counts in WoSCC as of March 6, 2026. These highly influential papers collectively received 7,681 (mean, 768.1; range, 254–2,967) citations, reflecting their pivotal role in shaping the field of pediatric AI research. The citation landscape reveals three distinct thematic domains. Medical imaging and diagnosis dominates the list, with Kermany et al.’s foundational work on image-based deep learning for disease diagnosis ranking first (2,967 citations) (9), followed by Brown et al.’s automated retinopathy of prematurity (ROP) diagnosis system (445 citations) (14) and Liang et al.’s transfer learning approach for pediatric pneumonia (261 citations) (17). Developmental and behavioral pediatrics is represented by Hazlett et al.’s neuroimaging study on early autism detection (772 citations) (11) and Barrett et al.’s methodological work on emotional expression analysis (1,124 citations) (10). Digital health and AI methodology comprises Park et al.’s guide for evaluating clinical AI performance (607 citations) (13), Torous et al.’s review of digital psychiatry (638 citations) (12), Lee et al.’s automated bone age assessment system (316 citations) (15), and Kaissis et al.’s privacy-preserving deep learning framework (295 citations) (16). Notably, Sinha et al.’s review on molecular sepsis diagnosis (254 citations) (18) represents the emerging intersection of AI and infectious disease diagnostics in pediatrics. The concentration of imaging studies among highly cited papers aligns with the broader bibliometric patterns in our dataset. The presence of general AI methodology papers in the top five underscores how foundational advances in medical AI have been subsequently adapted for pediatric applications. The diversity of topics—from ROP and pneumonia to autism and bone age—reflects the broadening scope of AI applications across pediatric subspecialties.

Figure 2 Annual publication volume of research papers related to the application of AI in pediatric diseases from 2016 to 2025. AI, artificial intelligence.

Co-citation analysis of foundational references

Beyond the citation impact of retrieved publications, co-citation analysis identifies the core references most frequently cited by the 1,893 articles in our dataset (Table 2). The most prominent thematic area, focused on ROP, comprises four studies that collectively demonstrate the maturity of AI for this condition. The highly co-cited work by Brown et al. established a benchmark by developing a deep convolutional neural network for diagnosing plus disease that outperformed most human experts (13). This was complemented by Wang et al. (21), who developed the DeepROP system for automated screening, and Taylor et al. (24), who created a quantitative severity score for monitoring disease progression. The inclusion of the International Classification of Retinopathy of Prematurity, 3^rd edition (ICROP3) international consensus statement by Chiang et al. further underscores how evolving AI capabilities are actively shaping clinical classification frameworks (20). A second significant thematic area addresses pediatric pneumonia diagnosis, featuring a foundational study by Kermany et al. that demonstrated a generalizable deep-learning framework validated on chest X-rays (9), and a later technical refinement by Liang et al. utilizing transfer learning with deep residual networks (17). Prenatal detection of congenital heart disease (CHD), a critical diagnostic challenge, is represented by the work of Arnaout et al., which demonstrated expert-level performance using an ensemble of neural networks on fetal ultrasound images (25). Notably, the list also includes one seminal methodological paper and two landmark adult-disease studies whose broad influence extends into pediatric research: Selvaraju et al.’s gradient-weighted class activation mapping (Grad-CAM) technique (19), which is pivotal for model explainability and trust, and the studies by Gulshan et al. (22) on diabetic retinopathy and Esteva et al. (23) on skin cancer, which, despite their adult focus, serve as crucial methodological and validation benchmarks that have profoundly informed the development paradigms for subsequent pediatric AI applications. In summary, the co-citation analysis highlights ROP as the most consolidated and high-impact area, with strong secondary foci on pediatric pneumonia and prenatal diagnosis, while also reflecting the field’s reliance on foundational advances in explainable AI and the influential paradigms pioneered in adult medical imaging research.

Core authors of research on the application of AI in pediatric diseases

Under attributed authorship, the 10 most productive authors on artificial-intelligence applications in pediatric diseases in the WoSCC [2016–2025] are ranked in Table 3. The USA institutions dominate the list, occupying the first six positions and accounting for 74.07% of the 81 articles credited to these top authors. J. Peter Campbell and Michael F. Chiang share the lead with 12 papers each (0.63%), followed by R. V. Paul Chan (11; 0.58%). Jayashree Kalpathy-Cramer contributes 10 publications (0.53%), ranking fourth. Beyond the USA, Udyavara Rajendra Acharya (Singapore), Fadi Thabtah (New Zealand), Tobias Banaschewski (Germany), and Quique Bassat (Spain) represent the remaining four nations, each with five or six papers, indicating that while North American researchers drive the topic’s output, individual hubs in Europe and Asia-Pacific are also emerging.

Research institutions engaged in the application of AI to pediatric diseases

Table 3 shows that Harvard University (94 articles, 4.97%), University of London (85 articles, 4.49%), and University of California System (81 articles, 4.28%) are the three most prolific institutions, together accounting for 13.73% of the 1,893 publications. North American universities occupy six of the top 10 ranks, with the remainder distributed across the UK, France, and China; no other organisation exceeded 50 articles. As depicted in Figure 3, the collaboration network (n=366; e=1,721; density =0.0258) reveals low connectivity. The largest connected component encompasses 294 institutions, accounting for 80% of the network, yet only 1.0% of the nodes are labeled, suggesting a low level of centrality for most institutions. Only McGill University (centrality =0.15), University of California System (centrality =0.14), and Brigham & Women’s Hospital (centrality =0.10) act as structural bridges, and none of the top-three producers attain similarly high centrality. Consequently, research on AI applications in pediatric diseases remains institutionally fragmented, with limited inter-institutional synergy.

Figure 3 Institutional collaboration network in pediatric AI research [2016–2025]. Node size represents publication volume; line thickness represents collaboration intensity. AI, artificial intelligence.

National or regional collaboration network analysis

In the analysis of the co-occurrence network of national or regional collaboration, a total of 113 nodes and 751 links were generated, with a network density of 0.1187 (see Figure 4). The USA, China, and England ranked first, second, and third in terms of publication output, with 659, 461, and 180 publications, respectively. The countries ranking 4^th to 10^th were Italy, Germany, Canada, India, Spain, South Korea, and Australia, in that order. The distribution of countries indicates that research in this field is predominantly concentrated in developed countries in Europe and America, with the USA occupying a leading position. France had the highest centrality (0.19), despite ranking 11^th in publication output. This suggests that France has extensive connections with other countries in the field of AI applications in pediatric diseases. Based on a comprehensive evaluation of publication output, centrality, the number of connections among research institutions, and the emergence of these institutions, the USA, China, and France have been identified as the most influential countries in the field of AI applications in pediatric diseases. These countries hold a dominant position in this domain. England and Spain both had a centrality of 0.17, while Germany and Brazil also exhibited relatively high centrality values of 0.13 and 0.10, respectively.

Figure 4 National or regional collaboration network in pediatric AI research [2016–2025]. Node size represents publication volume; line thickness represents collaboration intensity. AI, artificial intelligence.

Keyword co-occurrence analysis: frequency and centrality

Table 4 presents the top 10 keywords ranked by frequency of occurrence and centrality from 2016 to 2025. In terms of frequency, the most commonly occurring keywords are “machine learning” (rank 1, count 639), “children” (rank 2, count 388), and “artificial intelligence” (rank 3, count 313). These keywords, along with “diagnosis” (rank 4, count 254) and “deep learning” (rank 5, count 234), highlight the prominence of advanced computational techniques and their application in pediatric disease research. The keywords “classification” (rank 6, count 183), “adolescents” (rank 7, count 179), “risk” (rank 8, count 134), “prevalence” (rank 9, count 99), and “infants” (rank 10, count 91) further emphasize the focus on specific pediatric populations and key research themes. Regarding centrality, which indicates the connectivity of keywords within the research network, “classification” (rank 1, centrality 0.1) stands out as the most central keyword. Other keywords with high centrality include “autism spectrum disorder” (rank 2, centrality 0.09), “therapy” (rank 3, centrality 0.08), “risk factors” (rank 4, centrality 0.07), and “care” (rank 5, centrality 0.07). These terms, along with “patterns” (rank 6, centrality 0.07) and “abnormality” (rank 7, centrality 0.06), reflect the interconnectedness of research topics related to specific conditions and interventions. Overall, the high frequency of keywords related to machine learning and AI underscores the significant role of these technologies in pediatric disease research. The centrality of terms like “classification” and “autism spectrum disorder” highlights the importance of these topics in connecting various research areas within the field.

Keyword clustering analysis

Using the LLR algorithm, the software was employed to cluster keywords from the WoSCC database over the past decade. This analysis identified research frontiers by labeling and ranking each cluster, resulting in a network comprising 494 nodes, 2746 links, and a network density of 0.0226. Nine distinct clusters were delineated, as depicted in Figure 5. Specifically, cluster 0 (“depression”) and cluster 4 (“autism spectrum disorder”) highlight the significant attention given to mental health disorders in pediatrics, emphasizing the role of AI in diagnostics and therapeutic interventions. Cluster 1 (“deep learning”), cluster 5 (“artificial intelligence”), and cluster 7 (“machine learning”) collectively underscore the broad application of AI technologies in enhancing predictive models and diagnostic accuracy across various pediatric conditions. Meanwhile, cluster 2 (“predictive model”) and cluster 6 (“prediction model”) demonstrate a concentrated effort in developing and refining predictive analytics to improve patient outcomes. Cluster 3 (“inflammatory bowel disease”) points towards specific chronic conditions, indicating targeted research efforts in managing such diseases with the aid of AI. Lastly, cluster 8 (“model”) suggests a general focus on modeling techniques as a fundamental aspect of AI research in pediatrics. These research directions collectively provide innovative perspectives and solutions for addressing complex issues such as early diagnosis, personalized treatment, and improved management of both mental health disorders and chronic inflammatory conditions in pediatric patients. The identified clusters reflect a multifaceted approach to leveraging AI to advance pediatric healthcare.

Figure 5 Keyword cluster analysis in pediatric AI research [2016–2025]. Nine distinct clusters were identified using the LLR algorithm. AI, artificial intelligence; LLR, log-likelihood ratio.

Emerging trends in AI applications in pediatric research

CiteSpace offers burst detection to explore instances where citation counts undergo significant changes within a specific period (26). Figure 6 presents the top 10 keywords with the highest burst strength, where the red bars indicate a sudden increase in citation frequency during the corresponding timeframes. Over the last six years, keywords that have shown a notable surge in citation counts starting from 2020 include “virtual reality”, “growth”, “neural network”, and “respiratory distress syndrome”. These keywords suggest that research in AI in pediatric research has been increasingly focused on leveraging emerging technologies to enhance pediatric healthcare. The increased attention to “virtual reality” indicates a growing interest in using immersive technologies for therapeutic and diagnostic purposes in children. “Growth” reflects a focus on the application of AI in monitoring and predicting pediatric growth trajectories, which is crucial for early detection of developmental issues. The surge in “neural network” citations points to the expanding role of AI in complex data analysis and pattern recognition within pediatric research. Lastly, the rise in “respiratory distress syndrome” citations suggests a concentrated effort to apply AI in understanding and managing this critical condition in newborns. These trends highlight the ongoing exploration into how AI can be harnessed to address specific challenges in pediatric medicine, particularly in areas that require sophisticated data analysis and innovative therapeutic approaches. Researchers are actively investigating how AI can contribute to improving outcomes in these critical areas of pediatric health.

Figure 6 Timeline of burst keywords in pediatric AI research [2016–2025]. Red bars indicate periods of significant increase in citation frequency. AI, artificial intelligence.

Evolution of AI applications in pediatric research: a co-citation analysis

A co-citation cluster analysis was conducted on the cited references of literature from the WoSCC database over the past decade, resulting in a network consisting of 580 nodes, 1,722 links, and a network density of 0.0103. By selecting “Show top 50.0% paths”, 13 clusters were formed, as depicted in Figure 7. Figure 7 illustrates a timeline view of the co-citation network, showcasing the evolutionary process of research in the field of the application of AI to Pediatric Diseases, highlighting influential literature within specific time zones. Cluster #8 (autism diagnosis), cluster #5 (retinopathy of prematurity), cluster #7 (autism), and cluster #11 (bone age assessment) have the earliest publication years, with 2014, 2017, 2017, and 2017, respectively. The most recent publications are found in cluster #4 (autism spectrum disorder), cluster #9 (congenital heart disease), cluster #17 (MRI coupling), and cluster #10 (attention-deficit/hyperactivity disorder), published in 2021, 2021, 2021, and 2022, respectively.

Figure 7 Timeline of co-citation clusters in pediatric AI research [2016–2025]. Clusters are arranged chronologically by the year of their most cited reference. AI, artificial intelligence.

Discussion

This bibliometric study maps the global research landscape of AI applications in pediatric medicine over the past decade. The results reveal an almost 10-fold increase in annual publications (from 53 in 2016 to 526 in 2025, R²=0.93). The exponential growth curve conforms to the “exogenous shock” model of science expansion: B. Koçak et al. documented an exponential growth pattern in medical imaging AI, with an estimated annual growth rate of 29.8% and a publication doubling time of just 2.7 years (27). C. Langlotz et al. [2019] confirm a significant National Institutes of Health (NIH) workshop in 2018 that set research priorities for medical AI (28), while L. Martí-Bonmatí et al. highlight concurrent European efforts like the PRIMAGE (PRedictive In silico Multiscale Analytics to support cancer personalized diaGnosis and prognosis, Empowered by imaging biomarkers) project for pediatric imaging (29). Such policy-level infusions enlarge the carrying capacity of the literature system, explaining the 0.93 fit better than linear organic growth.

The citation impact analysis reveals the field’s intellectual foundations and influential directions. The top-cited publication, Kermany et al.’s Cell paper on image-based deep learning (2,967 citations) (9), demonstrates how generalizable AI frameworks developed for medical diagnosis have profoundly shaped pediatric applications. Notably, this study was conducted on adult populations but has been widely adapted for pediatric use, reflecting methodological precedent transfer from adult to pediatric medicine. Among truly pediatric-focused studies, Hazlett et al.’s neuroimaging work on early autism detection (772 citations) (11) and Brown et al.’s ROP diagnosis system (445 citations) (14) represent high-impact applications where AI has achieved clinical translation. The presence of two methodological guides—Barrett et al.’s critical appraisal of emotion research (1,124 citations) (10) and Park et al.’s evaluation framework for clinical AI (607 citations) (13)—underscores the field’s ongoing efforts to establish rigorous standards for AI validation in pediatric populations. Three thematic domains emerge from the highly cited literature. Medical imaging dominates, with applications spanning ROP, pneumonia, and bone age assessment. Developmental and behavioral pediatrics is represented by autism neuroimaging and emotion recognition research. Digital health infrastructure includes privacy-preserving methods and digital psychiatry platforms. This distribution reflects fundamental structural advantages of imaging data—high-contrast, standardized, two-dimensional inputs ideally suited to deep learning, with mature annotation protocols and publicly available datasets—rather than relative clinical importance. In contrast, other critical pediatric domains, including oncology, rare diseases, critical care decision support, pharmacovigilance, and genomics, face methodological barriers such as heterogeneous phenotypes, subjective assessments, limited training data, and privacy constraints, resulting in lower bibliometric visibility despite substantial clinical need.

Beyond the citation impact of retrieved publications, co-citation analysis identifies the core references most frequently cited by the 1,893 articles in our dataset. While AI studies demonstrate high diagnostic accuracy for ROP in controlled settings (14,20,21,24), real-world implementation remains limited by algorithm generalizability, lack of standardized data acquisition protocols, and insufficient external validation (30). A notable exception is an Indian telemedicine program achieving 100% sensitivity and 78% specificity for treatment-requiring ROP (31), suggesting telemedicine as a promising deployment pathway. For pediatric pneumonia, emerging studies demonstrate proof-of-concept clinical translation, though implementation remains limited. A prospective pilot in Nigeria achieved 58% external validation accuracy on radiologist-labeled pediatric chest radiographs, substantially lower than 86% internal accuracy, revealing performance gaps across healthcare environments (32). However, among over 40 certified AI products for chest imaging, few are specifically designed for pediatric populations (33), limiting pediatric-specific clinical translation. For prenatal CHD detection, clinical translation remains sparse despite promising research performance. Only limited examples of mainstream adoption exist, such as fetal ultrasound-based AI for prenatal detection (34). However, most AI initiatives for pediatric CHD remain at the prototype stage, with limited multi-center validation and significant underutilization in clinical settings (35,36). Barriers include regulatory approval requirements, clinician acceptance, data privacy concerns, and the need for prospective validation studies (14,37,38). Notably, bone age assessment represents a relatively mature exception, with commercially available AI software in several countries, though comprehensive regulatory approval data for pediatric-specific AI products remain limited. In contrast, ROP AI has no formally approved commercial products globally; pediatric pneumonia AI has some approved commercial products spanning diverse technologies, though pediatric-specific applications remain limited; and CHD AI is concentrated in prenatal screening and select postnatal applications rather than widespread pediatric deployment. This pattern reveals that high bibliometric activity in these domains primarily reflects imaging validation studies rather than uniform real-world adoption, with substantial variation across disease areas.

The divergence between Table 1 and Table 2 reveals ROP’s role as a methodological anchor: while generalizable AI frameworks attract higher direct citations, ROP studies dominate co-citation references (4 of 10), reflecting their foundational status for validation. This concentration stems from temporal asymmetry in dataset availability: the imaging-ROP (i-ROP) cohort expanded from 77 images in 2015 (39) to 5,511 photographs by 2018 (14), establishing early standardized benchmarks that pediatric pneumonia imaging lacked. In contrast, comparable pediatric chest X-ray datasets emerged only after 2020, delaying analogous consolidation. This infrastructure gap, verifiable through i-ROP’s documented expansion, explains ROP’s disproportionate co-citation influence as a methodological precedent, despite similar clinical importance across these imaging domains.

Our clustering results corroborate Ganatra’s assertion that AI has moved from proof-of-concept to early disease detection in pediatrics (2), but they also refine the timeline: the co-citation burst of Kermany et al.’s pneumonia paper peaked in 2020 and has already declined (9), suggesting that diagnostic chest-X-ray models are transitioning from “hot topics” to “textbook knowledge”. Unexpectedly, two adult-focused studies—Gulshan’s diabetic retinopathy classifier (22) and Esteva’s skin-cancer convolutional neural network (CNN) (23)—rank among the top 10 most co-cited references. This cross-domain citation reflects methodological precedent transfer from adult to pediatric applications: once an algorithmic architecture gains clinical credibility in adults, its adaptations are rapidly imported into pediatric equivalents, shortening the validation cycle but amplifying algorithmic bias when demographic spectra diverge. The phenomenon connects bibliometric patterns to real-world regulatory dynamics and clinical deployment risks.

Beyond citation counts, the micro-structure of authorship and national affiliation reveals how knowledge actually travels. The author-level network exhibits a “high-productivity/low-inter-institutional” pattern: the six most prolific scientists are all based in the USA (Table 3), yet none appear among the three structural-bridge institutions listed in the institutional network (McGill, University of California System, and Brigham & Women’s Hospital). This mismatch suggests that domestic, intra-hospital teams generate volume but rarely serve as gateways between otherwise disconnected clusters, reinforcing the globally low network density (0.0258) reported above. Second, France ranks only 11^th in output (128 papers) but holds the highest betweenness centrality (0.19) in the country co-occurrence network. This disconnect between productivity and connectivity reflects France’s structural position in European health research networks, where it has been identified as a pivotal collaborator alongside Germany, the United Kingdom, the Netherlands, and Switzerland (40). However, this brokerage role likely draws on historical rather than emerging strengths: France-Africa biomedical partnerships [2012–2021] reveal a “constant decline” in research output and limited diversification beyond infectious diseases (41), suggesting that France’s centrality in pediatric AI represents residual structural capital from established European Union (EU) frameworks rather than active domain expansion. Taken together, these relational disparities underscore that in pediatric-AI research, historical connectivity persists even as volume leadership shifts, complicating productivity-centric assessments of global influence (42,43).

Despite rapid publication growth, our analysis reveals substantial gaps in the evidence base required for safe clinical translation. Regulatory science remains underdeveloped: few pediatric-specific AI devices have received Food and Drug Administration (FDA) approval or Conformité Européenne (CE) marking compared to adult applications, and standardized evaluation frameworks for pediatric populations are lacking (44). Implementation research is scarce, with limited studies examining workflow integration, cost-effectiveness, or clinician acceptance in real-world pediatric settings. Equity concerns are pronounced: the geographic concentration of research in high-income countries and underrepresentation of diverse pediatric populations in training datasets raise concerns about algorithmic bias and generalizability to underserved communities. Most critically, prospective trials are notably absent from the highly cited literature, with the field relying predominantly on retrospective validation studies rather than randomized controlled trials or prospective cohort studies. Additional barriers to translation include data heterogeneity across pediatric age groups and conditions, the scarcity of large, diverse pediatric-specific datasets compared to adult medicine, and dataset limitations stemming from ethical constraints on data sharing and the challenges of collecting representative samples across developmental stages (2,45). These gaps and barriers collectively hinder the transition from bibliometric activity to evidence-based clinical practice.

This study has several limitations. Firstly, the bibliometric analysis is based on publications indexed in the WoSCC database, which may not capture all relevant research, particularly from non-indexed journals or grey literature. Secondly, the analysis focuses on the past decade, potentially omitting foundational work from earlier years. Thirdly, the study relies on citation data, which may not fully reflect the practical impact of AI applications in clinical settings. Fourthly, our search strategy required disease-related terms (“disease*”, “disorder*”, “illness*”, and “syndrome*”), which may have excluded AI studies in imaging, screening, developmental assessment, or procedural guidance that do not explicitly use these terms. While this approach ensured focus on clinical disease applications, it may have limited capture of broader pediatric AI research. Fifthly, the restriction to WoSCC excludes Scopus, PubMed, Embase, IEEE Xplore, and preprint servers, which may have limited capture of emerging AI research, particularly from computer science and engineering perspectives. Sixth, the analysis relies on CiteSpace as a single analytical tool without cross-validation using alternative software. While CiteSpace is widely validated for bibliometric visualization, future studies should confirm the robustness of network structures and clustering solutions across multiple platforms. Future research should address these limitations by incorporating a broader range of databases and conducting longitudinal studies to assess the long-term impact of AI applications. Additionally, further research should focus on developing explainable AI models to enhance clinical trust and adoption, as well as exploring the ethical implications of AI in pediatric healthcare.

Conclusions

AI applications in pediatrics are expanding rapidly, but translation pathways differ by domain maturity. For mature imaging domains such as ROP, characterized by high co-citation consolidation and established public datasets, priorities should shift from algorithm development to multi-center prospective validation and regulatory science. For high-volume but externally inconsistent domains such as pediatric pneumonia, where prospective pilots reveal performance gaps across healthcare environments, data standardization and cross-environment validation should precede deployment. For emerging areas identified by keyword burst detection, including virtual reality, growth analytics, and neonatal respiratory distress syndrome, ethical frameworks and prospective data collection protocols are urgently needed. Transnational consortia should prioritize evidence-scarce domains where our analysis reveals low network density and limited inter-institutional synergy. These differentiated strategies, derived from bibliometric indicators of field maturity, validation consistency, and collaboration structure, underscore that connectivity rather than productivity drives global influence in pediatric AI research.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the BIBLIO reporting checklist. Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0116/rc

Peer Review File: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0116/prf

Funding: This work was supported by the Sichuan Natural Science Foundation: Research on Hybrid Network Methods for Revealing Scientific Structure and Evolution (No. 2024NSFSC1064).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0116/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Ghunaim L, Ahmed SA, Agha A, et al. The future of pediatric care: AI and ML as catalysts for change in genetic syndrome management. J Med J 2024;58:510-28.
2. Ganatra HA. Machine Learning in Pediatric Healthcare: Current Trends, Challenges, and Future Directions. J Clin Med 2025;14:807. [Crossref] [PubMed]
3. Kerth JL, Hagemeister M, Bischops AC, et al. Artificial intelligence in the care of children and adolescents with chronic diseases: a systematic review. Eur J Pediatr 2024;184:83. [Crossref] [PubMed]
4. Carroll AE, Christakis DA. Call for papers on artificial intelligence applied to pediatric care. JAMA Pediatr 2023;177:884-5.
5. Shu B, Zhang S, Gao J, et al. The digital transformation and future era: bibliometric view of artificial intelligence application in pediatric surgery. Front Pediatr 2025;13:1528666. [Crossref] [PubMed]
6. Wang J, Liang Y, Cao S, et al. Application of Artificial Intelligence in Geriatric Care: Bibliometric Analysis. J Med Internet Res 2023;25:e46014. [Crossref] [PubMed]
7. Velasco Cardona DC, Cardona-Acevedo S, Valencia-Arias A, et al. Bibliometric analysis of the application of artificial intelligence techniques in bacteriology: a PRISMA-guided research agenda. Front Microbiol 2025;16:1641967. [Crossref] [PubMed]
8. Wei Y, Mei Z, Xie C, et al. Mapping the application landscape of artificial intelligence in prostate cancer: a global bibliometric analysis. Int J Surg 2026;112:2988-3002. [Crossref] [PubMed]
9. Kermany DS, Goldbaum M, Cai W, et al. Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning. Cell 2018;172:1122-1131.e9. [Crossref] [PubMed]
10. Barrett LF, Adolphs R, Marsella S, et al. Emotional Expressions Reconsidered: Challenges to Inferring Emotion From Human Facial Movements. Psychol Sci Public Interest 2019;20:1-68. [Crossref] [PubMed]
11. Hazlett HC, Gu H, Munsell BC, et al. Early brain development in infants at high risk for autism spectrum disorder. Nature 2017;542:348-51. [Crossref] [PubMed]
12. Torous J, Bucci S, Bell IH, et al. The growing field of digital psychiatry: current evidence and the future of apps, social media, chatbots, and virtual reality. World Psychiatry 2021;20:318-35. [Crossref] [PubMed]
13. Park SH, Han K. Methodologic Guide for Evaluating Clinical Performance and Effect of Artificial Intelligence Technology for Medical Diagnosis and Prediction. Radiology 2018;286:800-9. [Crossref] [PubMed]
14. Brown JM, Campbell JP, Beers A, et al. Automated Diagnosis of Plus Disease in Retinopathy of Prematurity Using Deep Convolutional Neural Networks. JAMA Ophthalmol 2018;136:803-10. [Crossref] [PubMed]
15. Lee H, Tajmir S, Lee J, et al. Fully Automated Deep Learning System for Bone Age Assessment. J Digit Imaging 2017;30:427-41. [Crossref] [PubMed]
16. Kaissis G, Ziller A, Passerat-Palmbach J, et al. End-to-end privacy preserving deep learning on multi-institutional medical imaging. Nat Mach Intell 2021;3:473-84.
17. Liang G, Zheng L. A transfer learning method with deep residual network for pediatric pneumonia diagnosis. Comput Methods Programs Biomed 2020;187:104964. [Crossref] [PubMed]
18. Sinha M, Jupe J, Mack H, et al. Emerging Technologies for Molecular Diagnosis of Sepsis. Clin Microbiol Rev 2018;31:e00089-17. [Crossref] [PubMed]
19. Selvaraju RR, Cogswell M, Das A, et al. Grad-CAM: visual explanations from deep networks via gradient-based localization. Int J Comput Vis 2020;128:336-59.
20. Chiang MF, Quinn GE, Fielder AR, et al. International Classification of Retinopathy of Prematurity, Third Edition. Ophthalmology 2021;128:e51-68.
21. Wang J, Ju R, Chen Y, et al. Automated retinopathy of prematurity screening using deep neural networks. EBioMedicine 2018;35:361-8. [Crossref] [PubMed]
22. Gulshan V, Peng L, Coram M, et al. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs. JAMA 2016;316:2402-10. [Crossref] [PubMed]
23. Esteva A, Kuprel B, Novoa RA, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature 2017;542:115-8. Erratum in: Nature 2017;546:686.
24. Taylor S, Brown JM, Gupta K, et al. Monitoring Disease Progression With a Quantitative Severity Scale for Retinopathy of Prematurity Using Deep Learning. JAMA Ophthalmol 2019;137:1022-8. [Crossref] [PubMed]
25. Arnaout R, Curran L, Zhao Y, et al. An ensemble of neural networks provides expert-level prenatal detection of complex congenital heart disease. Nat Med 2021;27:882-91. [Crossref] [PubMed]
26. Chen C. CiteSpace II: detecting and visualizing emerging trends and transient patterns in scientific literature. J Am Soc Inf Sci Technol 2006;57:359-77. [Crossref]
27. Kocak B, Baessler B, Cuocolo R, et al. Trends and statistics of artificial intelligence and radiomics research in Radiology, Nuclear Medicine, and Medical Imaging: bibliometric analysis. Eur Radiol 2023;33:7542-55. [Crossref] [PubMed]
28. Langlotz CP, Allen B, Erickson BJ, et al. A Roadmap for Foundational Research on Artificial Intelligence in Medical Imaging: From the 2018 NIH/RSNA/ACR/The Academy Workshop. Radiology 2019;291:781-91. [Crossref] [PubMed]
29. Martí-Bonmatí L, Alberich-Bayarri Á, Ladenstein R, et al. PRIMAGE project: predictive in silico multiscale analytics to support childhood cancer personalised evaluation empowered by imaging biomarkers. Eur Radiol Exp 2020;4:22. [Crossref] [PubMed]
30. Gensure RH, Chiang MF, Campbell JP. Artificial intelligence for retinopathy of prematurity. Curr Opin Ophthalmol 2020;31:312-7. [Crossref] [PubMed]
31. Campbell JP, Singh P, Redd TK, et al. Applications of Artificial Intelligence for Retinopathy of Prematurity Screening. Pediatrics 2021;147:e2020016618. [Crossref] [PubMed]
32. Togunwa TO, Babatunde AO, Fatade OE, et al. Detection of pneumonia in children through chest radiographs using artificial intelligence in a low-resource setting: A pilot study. PLOS Digit Health 2025;4:e0000713. [Crossref] [PubMed]
33. Schalekamp S, Klein WM, van Leeuwen KG. Current and emerging artificial intelligence applications in chest imaging: a pediatric perspective. Pediatr Radiol 2022;52:2120-30. [Crossref] [PubMed]
34. Sen S, Ramakrishnan S. Artificial intelligence in pediatric cardiology: Where do we stand in 2024? Ann Pediatr Cardiol 2024;17:93-6. [Crossref] [PubMed]
35. Van den Eynde J, Kutty S, Danford DA, et al. Artificial intelligence in pediatric cardiology: taking baby steps in the big world of data. Curr Opin Cardiol 2022;37:130-6. [Crossref] [PubMed]
36. Jone PN, Gearhart A, Lei H, et al. Artificial Intelligence in Congenital Heart Disease: Current State and Prospects. JACC Adv 2022;1:100153. [Crossref] [PubMed]
37. Suha KT, Lubenow H, Soria-Zurita S, et al. The Artificial Intelligence-Enhanced Echocardiographic Detection of Congenital Heart Defects in the Fetus: A Mini-Review. Medicina (Kaunas) 2025;61:561. [Crossref] [PubMed]
38. Campbell JP, Chiang MF, Chen JS, et al. Artificial Intelligence for Retinopathy of Prematurity: Validation of a Vascular Severity Scale against International Expert Diagnosis. Ophthalmology 2022;129:e69-76. [Crossref] [PubMed]
39. Ataer-Cansizoglu E, Bolon-Canedo V, Campbell JP, et al. Computer-Based Image Analysis for Plus Disease Diagnosis in Retinopathy of Prematurity: Performance of the “i-ROP” System and Image Features Associated With Expert Diagnosis. Transl Vis Sci Technol 2015;4:5. [Crossref] [PubMed]
40. Tapia ECO, Albitres J, Calderón PCC, et al. Health and medical informatics research: Identifying international collaboration patterns at the country and institution level. Iberoamerican Journal of Science Measurement and Communication 2024;4:1-16.
41. Le Bihan E, Debré P, Migus A, et al. Analyse des limites du partenariat France–Afrique en recherche biomédicale à travers les données de la bibliométrie et des essais cliniques. Bulletin de l’Académie Nationale de Médecine 2024;208:326-37.
42. Vernooij RWM, Hockham C, Strippoli G, et al. Haemodiafiltration versus haemodialysis for kidney failure: an individual patient data meta-analysis of randomised controlled trials. Lancet 2024; Epub ahead of print. [Crossref]
43. Abubakar I, Plasencia A, Bärnighausen T, et al. Horizon Europe: towards a European agenda for global health research and innovation. Lancet 2019;393:1272-3. [Crossref] [PubMed]
44. Kelly BS, Lee J, Antram E, et al. Safe for kids? AI medical devices in radiology overlook paediatric suitability. Eur Radiol 2026;36:2083-90.
45. Park T, Lee IH, Lee SW, et al. Artificial intelligence in pediatric healthcare: current applications, potential, and implementation considerations. Clin Exp Pediatr 2025;68:641-51. [Crossref] [PubMed]