Research trends and hotspots in the applications of artificial intelligence in CT based on Web of Science—bibliometric research (2015–2024)

Introduction

In recent years, artificial intelligence (AI) has emerged as a research hotspot in the field due to the rapid advancement of intelligent medical imaging diagnosis and the increasing demand for analyzing complex medical images. Computed tomography (CT), as a discipline heavily reliant on digital systems and image analysis, occupies a relatively advantageous position in the application of AI (1,2). Over the past decade, AI for CT has undergone distinct developmental shifts since the breakthrough of ResNet in 2015 (3): early research relied on shallow machine learning for basic CT image processing, followed by a boom in U-Net and three-dimensional (3D) convolutional neural network (CNN)-based automatic lesion and organ segmentation from 2018 to 2020 (4), then gradual expansion toward adaptive radiotherapy, multi-modal fusion and real-world clinical verification between 2021 and 2023 (5), and most recently the rapid rise of generative models and medical foundation models in 2024 (6,7). Nevertheless, the whole field is confronted with prominent challenges, including a severe mismatch between surging publication volume and slow clinical translation, a relative shortage of high-quality clinical studies from China, unbalanced research progress between diagnostic CT and CT-guided radiotherapy, insufficient standardized shared CT datasets, and inadequate exploration of interpretable AI and health economic evaluation.

Bibliometric studies employ quantitative statistical methods to analyze literature metadata (e.g., keywords, citations, collaboration networks) and map the research landscape, identify trends, or reveal knowledge gaps, differing from the traditional reviews, which focus on qualitative synthesis and critical interpretation of research content to answer specific academic questions (8). This study focuses on CT research in order to use bibliometric analysis to examine the publication landscape of CT-specific studies versus multimodal imaging studies within the broader literature on AI in radiology, covering the period 2015–2024 and based on data indexed in the Web of Science Core Collection (WoSCC). The results could elucidate the developmental trends and research hotspots of AI in CT, providing a reference framework for further in-depth investigations in this domain. We present this article in accordance with the BIBLIO reporting checklist (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2026-1-0015/rc).

Methods

Data source and search strategy

This study’s data were retrieved from the WoSCC platform. A specialized search query was constructed as follows: (TS=(“AI”) OR TS=(“artificial intelligence”) OR TS=(“intelligent healthcare”)) AND (TS=(“medical image”) OR TS=(“medical imaging”) OR TS=(“medical imaging technology”) OR TS=(“CT”) OR TS=(“computed tomography”) OR TS=(“multi-detector row CT”) OR TS=(“MDCT”) OR TS=(“X-ray computed tomography”)) NOT (TS=(“Magnetic Resonance Imaging”) OR TS=(“MRI”) OR TS=(“Emission Computed Tomography”) OR TS=(“PET”) OR TS=(“Optical Tomography”) OR TS=(“radiotherapy”)). The time range for the search is from 2015-01-01 to 2024-12-31. We selected 2015 as the starting year because the invention of ResNet in this year realized the large-scale application of deep learning in CT diagnosis, forming the technical origin of this field. Publications before 2015 were scarce and limited to traditional image processing without an independent AI-CT research system. The 2015–2024 time span covers the complete developmental cycle of the field and ensures that bibliometric analysis focuses on core AI research. It accurately reflects the changes in research hotspots over a 10-year period.

Research methods

We retrieved a total of 6,097 publications from WoSCC, but only 4,384 were included after screening, and we checked these papers by three authors. A two-step literature evaluation framework was implemented for all retrieved records. Step 1: preliminary screening based on titles and abstracts. Three independent researchers screened all papers to remove irrelevant documents, such as pure CT radiotherapy research, articles without AI algorithms, and irrelevant imaging modalities. Disagreements were resolved through group discussion. Step 2: full-text evaluation for eligibility confirmation. We further assessed the core research content, study type, experimental design and research objects of the preliminarily retained articles. Review papers, editorial materials, short conference abstracts and case reports were excluded to avoid interference with bibliometric indicators. Only original research articles focusing on AI-aided diagnostic CT analysis were finally incorporated into the quantitative analysis dataset. The present study utilized VOSviewer version 1.6.20 to filter data, with Microsoft Excel 2019 employed for the summarization and systematic analysis of the selected papers. Bibliometric research and visual analysis were conducted from aspects such as annual publication trends, author collaboration trends, country distribution, institutional collaboration trends, journal distribution, high-frequency keywords, and co-cited literature. The flow chart of the literature screening is shown in Figure 1.

Figure 1 The flow chart of searching papers in databases. WoSCC, Web of Science Core Collection.

The search strategy summary is shown in Table 1.

Results

Annual publication trends

Prior to 2018, research on AI in radiology was relatively limited and at an early stage. From 2019 onwards, the publication volume increased markedly each year, indicating that the field had entered a rapid development phase. By 2024, the publication volume reached 1,338 articles, approximately 20 times that of 2018 (67 articles) and 10 times that of 2019 (138 articles). Publications from the past five years (2020–2024) account for 93.3% of the total publications during the entire period (2015–2024) (4,089/4,384). Overall, research on AI in medical imaging is currently in an explosive growth stage and exhibits broad prospects for future development, as depicted in Figure 2.

Figure 2 Numbers of publications (2015–2024).

International cooperation

Taking countries as analytical nodes, it constructed a national collaboration network comprising 113 nodes and 4,617 edges in Figure 3. Only a few countries conduct relatively independent research in this field without collaboration with other nations. The top 10 countries in terms of total number of publications are presented in Table 2. Regarding publication volume, the leading five countries in this field are China (n=1,146), the United States (n=1,093), Italy (n=360), South Korea (n=323), and India (n=281). Among these, the United States exhibits the highest total citation frequency (29,452 citations), while the United Kingdom attains the highest average citations per paper, approximately 34.39. Although China has the highest number of published articles, its total citations amount to 19,833, ranking second, while its average citation count is only 17.31, placing it seventh. This underscores the prominent positions of China and the United States in this domain, with the United Kingdom playing a central role in international collaboration. The cooperative relationships among countries, as shown in Figure 4, reveal close connections between different nations, with substantial collaboration within this field.

Figure 3 Map of international cooperation relations.

Figure 4 Map of international cooperation relations between the top 10 most productive countries.

Institutional cooperation

By setting the node type to ‘Institution’, the institutional collaboration network was constructed, involving 6,522 institutions in total, and the Visualized Map of Institutional Collaboration Analysis is shown in Figure 5. Among them, 4,427 institutions published only one paper, while the top ten institutions each published more than 50 papers. The top 10 institutions ranked by publication volume are listed in Table 3.

Figure 5 Visualized map of institutional collaboration analysis.

Between 2015 and 2024, the institution with the strongest collaborative link strength is Mayo Clin. Harvard Medical School ranks highest in publication volume, total citation frequency, and average citation frequency, identifying it as the most influential institution in the field during this period. The table further shows that half of the top 10 institutions are affiliated with China, while 4 institutions are affiliated with the USA, underscoring the significant contributions of both the United States and China to research in this area, consistent with the findings of the national collaboration analysis. Additionally, South Korea holds a position among the top 10 countries, which not only reflects their own capabilities but also highlights the crucial role played by research teams from countries with smaller populations.

Author collaboration trends

Statistical analysis indicates that this study involved a total of 35,611 participants. The team led by U. Joseph Schoepf published the highest number of papers, totaling 27 articles with 845 citations and an average citation count of 31.29 per paper, making them the most prolific researchers in this field. Ranking second were Summers and Ronald M (n=22), with 1,040 citations, followed by Saba and Luca (n=20), with 486 citations. These teams demonstrated their research prowess in this field through the number of published papers and total citations.

Counting the number of authors per article and using two indicators to measure collaboration: collaboration rate and collaboration level. Collaboration rate refers to the proportion of papers with two or more authors relative to the total number of documents. Collaboration level is generally expressed as the average author collaboration degree across all papers, where the number of authors on each paper defines that paper’s author collaboration degree (9).

As shown in Table 4, the literature demonstrates a high collaboration rate, consistently above 90% since 2015 and reaching 100% in both 2016 and 2017. The number of single-author papers remains persistently low, maybe because it is challenging for an individual researcher to publish more studies without close collaboration. The level of literature collaboration peaked at 8.9 in 2019, displaying a unimodal distribution. This coincided precisely with the onset of the coronavirus disease 2019 (COVID-19) pandemic. Therefore, we hypothesize that the progression of the pandemic may have hindered collaborative interactions among researchers.

Journal distribution

Using journal sources as the analytical node, a total of 1,128 journals were identified. As shown in Figure 6, 628 journals only published one paper, 77 journals published more than 10 articles, but only 4 journals published over 100 articles, namely Diagnostics (n=134), Scientific Reports (n=124), European Radiology (n=118), and IEEE Access (n=107), which are popular journals in the domain of AI applications in imaging research. Among the aforementioned four journals, the first three fall in the Journal Citation Reports (JCR) Q1, while the last one belongs to the JCR Q2. Their respective impact factors are 3.3, 3.9, 4.7, and 3.6. The journal with the highest citations is Cancer Research, boasting 4,012 citations, although the number of publications is only two. It belongs to JCR Q1, and the impact factor is 16.6.

Figure 6 Journals publication statistics.

High-frequency keywords

Centering on keyword occurrence frequencies to identify research hotspots of AI in Radiology, thereby providing a foundation for understanding the research landscape in this field. The frequency of keyword occurrences reflects current research hotspots, while the linkage strength between keywords indicates their significance. The results show that the included 4,384 literature comprises a total of 13,267 keywords, among which 29 keywords have frequencies greater than 100, and a visualization analysis of these high-frequency keywords is presented in Figure 7. Listing the 10 high-frequency keywords in Table 5, we found that the keyword with the highest occurrence frequency and linkage strength is “artificial intelligence”, appearing 1,685 times with a linkage strength of 3,161. The following most prominent keywords are “deep learning” and “machine learning”, with occurrence frequencies of 1,080 and 507, respectively. By excluding retrieval-limiting terms such as “artificial intelligence”, “CT”, and “computed tomography”, research hotspot topics primarily focus on “deep learning”, “machine learning”, “classification”, “COVID-19”, and “diagnosis”. After performing cluster analysis, the focus primarily lies on methods and general technologies, cardiovascular applications, oncology applications, and COVID-19-related fields. It is thus apparent that current research hotspots concentrate mainly on three dimensions: disease diagnosis, core technical methods, and key application tasks. These research directions intersect and collectively drive the rapid advancement of the field.

Figure 7 The high-frequency keywords network visualization (A) and keywords density visualization (B).

The trend of keyword changes over the years is shown in Figure 8. We observed that core technology-related keywords primarily include AI, deep learning, and machine learning—these three constitute the fundamental driving forces of the field. AI was only sporadically mentioned from 2015 to 2018, then saw rapid growth starting in 2019, reaching an annual peak of 461 mentions in 2024—making it the leading core keyword across all domains. Its overall trajectory reflects linear, exponential growth, underscoring AI as the overarching central concept throughout this research field. Deep learning has grown in tandem with AI, maintaining steady expansion since 2020 with increasing annual frequency—reaching 284 occurrences by 2024. As a core algorithm in AI subfields, it remains the second most researched area without significant decline. Machine learning has maintained steady growth since 2019, experiencing only a minor decline in 2022 before rebounding strongly in 2023–2024. As a traditional branch of machine learning, its popularity remains consistently high, serving as a complementary technical approach alongside deep learning. During the field’s development, the early stage (2015–2018) saw only CT-related keywords, with virtually no literature on AI or deep learning, and research primarily focused on traditional imaging; from 2019 to 2020, the field experienced rapid growth as keywords such as AI, deep learning, classification, and diagnosis emerged en masse, marking its official entry into the AI+CT era; the period from 2021 to 2024 represented the maturity phase, characterized by substantial simultaneous growth across all core keywords and coordinated advancements in technology, imaging, and clinical applications. The research initially focuses on fundamental algorithms (deep learning, machine learning), shifted to CT imaging modality development in the mid-stage, and gradually shifted toward clinical diagnostic applications in the later phase. Periodic pandemic-related hotspots merely caused short-term disturbances to the overall developmental trajectory, highlighting the distinction between long-term fundamental research and short-term event-driven focal points.

Figure 8 The trend of hot frequency keywords.

Co-citation analysis of literature

Highly cited literature serves as a primary indicator of an article’s influence within its discipline; a higher citation count typically signifies greater impact and underscores the work’s significance in the field. An analysis of the 20 most frequently cited articles between 2015 and 2024 revealed that these articles have a minimum citation count of 278 and seven publications focused on the application of AI technologies in detecting the COVID-19. These studies included efficiency improvements achieved through automatic detection of new models, as well as a distinct differentiation between pneumonia caused by COVID-19 and non-COVID-19 etiologies. This highlights the relevance of imaging examinations as adjunct diagnostic tools, whose ultimate purpose is to serve clinical practice, address current challenges, and drive meaningful progress in medical science (10-16).

The article with the highest citation count [3,724] originates from the team led by van Griethuysen (17), which developed the open-source PyRadiomics platform to address the lack of standardized algorithm definitions and image processing in radiomics and AI technologies and the article elaborated on its workflow and framework and demonstrates its application in pulmonary lesion feature analysis, thereby providing new perspectives for the further integration of radiomics with AI. Gupta (18) and Mahmud (19), the teams conducted a comprehensive and up-to-date review of deep learning, reinforcement learning, and deep reinforcement learning technologies in health informatics and their applications in biological data mining. Leveraging cross-disciplinary dataset comparisons, the authors assessed various deep learning methodologies, offering a critical analysis of their relative merits, limitations, and developmental trajectories. In doing so, they delineated the prevailing open issues in this cutting-edge domain and charted a course for future research. Xu (20) demonstrated that deep learning can integrate imaging scans from multiple time points to enhance the accuracy of clinical outcome prediction. Yang (21) and Giger (22) conducted novel research in the field of machine learning. In clinical applications, Kermany’s (23) team developed diagnostic tools based on a deep learning framework to screen for common treatable blinding retinal diseases and to diagnose pediatric pneumonia, thereby facilitating early treatment and improving clinical outcomes. Kim’s team (24) investigated the application of deep CNNs in fracture detection. In addition, related literature has explored the application of radiomics in cancer immunotherapy and in predicting distinct radiological phenotypes driven by somatic tumor mutations (25,26).

The foregoing analysis indicates that AI has extensive applications in medicine; however, the focus remains primarily on deep learning, radiomics, and related areas, progressing from theoretical frameworks to practical implementations.

Discussion

With the rapid advancement of AI, which has been widely applied across various sectors, we are inevitably moving towards a smarter and more information-driven era. As medicine is a vital component of society dedicated to safeguarding human life and health, it must also progress continuously and embrace the characteristics of the times.

AI has emerged as a transformative force in CT imaging, revolutionizing virtually every aspect of the CT workflow—from image acquisition and reconstruction to lesion detection, segmentation, diagnosis, and clinical decision support (27,28). Deep learning algorithms, particularly CNNs, have demonstrated remarkable capabilities in reducing radiation dose while maintaining or enhancing image quality through advanced reconstruction and denoising techniques (29,30). AI-powered tools enable automatic segmentation of anatomical structures, accurate detection and characterization of pulmonary nodules (31), coronary artery plaques (32), intracranial hemorrhage (33), and a broad spectrum of pathological lesions across multiple organ systems. Beyond pure image analysis, AI integrates radiomics features with clinical data to support prognosis prediction, treatment response assessment, and patient risk stratification (34). Despite challenges related to data heterogeneity, model interpretability, and regulatory approval, AI is poised to redefine precision medicine through its seamless integration with CT imaging.

In this context, this study retrieved relevant literature on the application of AI in CT medical imaging from the WoSCC core database over the past decade and conducted visual analysis using VOSviewer. From the perspectives of annual publication trends, author collaboration trends, country distribution, institutional collaboration trends, journal distribution, high-frequency keywords, and co-cited literature, a comprehensive understanding of the current research status of AI applications in CT worldwide has been gained. And the residency and affiliations of the authors were not considered for the evaluation of international cooperation.

A total of 4,384 papers were included in this study. The number of published papers has been increasing annually. Especially against the backdrop of the COVID-19 pandemic, the application of AI in CT has received substantial support and advancement, and now the research in the field is in a period of explosive development. At present, the top five countries by the number of articles published in this field are China, the United States, Italy, South Korea, and India. Among them, the United States has the highest total citation frequency, and the United Kingdom has the highest average citation per article. Although China publishes the largest number of articles, its average citation frequency does not perform equally well. The reason may be that while China conducts more research in this field, it produces relatively fewer high-quality articles. This also reminds Chinese researchers to place greater emphasis on quality in future studies and enhance their international influence through in-depth research. The institution with the highest number of published papers, total citation frequency, and average citation frequency was Harvard Medical School, which was the most influential institution in related research during this period. The results also show that some countries have conducted in-depth research in this field, while many others are still in the exploratory phase; therefore, countries and institutions should strengthen cooperation to develop new applications and promote medical development.

The research involved a total of 1,128 journals. Of these, 628 journals only published one paper, but only 4 journals published more than 100 articles—specifically, Diagnostics, Scientific Reports, European Radiology, and IEEE Access. The above are popular journals in the field, however, none of them are the most cited journals. The reason may be that these journals better meet authors’ submission requirements, whereas high-impact journals with numerous citations, despite their greater credibility, pose significant challenges for publication.

The research hotspots mainly focus on aspects such as “deep learning”, “machine learning”, “classification”, “COVID-19”, and “diagnosis”. Categorized by theme, the focus is primarily on methodologies and general technologies, cardiovascular applications, oncology applications, and COVID-19-related research. Research on AI-assisted diagnosis of COVID-19 has advanced rapidly during this phase, likely due to the swift progression of the pandemic. Researchers developed numerous AI tools to address the complex and dynamic clinical manifestations and alleviate the workload of physicians. The article with the highest citation rate was published by van Griethuysen’s team in Cancer Research (2017) (17).

According to our dataset indicators covering the proportion of highly cited papers, publications in top radiology journals and cross-border collaboration intensity, China produces the largest number of relevant articles but lacks high-quality research outputs. Supported by global AI imaging literature, this gap arises from three structural factors: domestic academic evaluation mechanisms prioritize publication quantity and favour short-cycle incremental algorithm improvements over long-term multi-center prospective trials with solid clinical evidence (35); developed countries maintain stably funded integrated medical-engineering research platforms, whereas domestic interdisciplinary cooperation mostly relies on temporary projects without permanent joint teams (36); standardized multi-disease public CT imaging datasets are insufficient domestically, while Western scholars follow unified data annotation and reporting specifications to facilitate rigorous clinical verification (37,38). Together, these structural discrepancies lead to fewer high-impact clinical AI-CT studies from China relative to Western leading countries.

Keyword analysis shows that while methodological papers dominate the literature, terms related to clinical validation and regulatory approval are rare, highlighting a significant gap between academic output and clinical translation. This disconnect stems from three main barriers (39): technically, models trained on limited single-site datasets lack generalizability; clinically, weak medical-engineering collaboration leads developers to prioritize metrics over real-world workflow needs; and institutionally, inadequate regulatory standards and reimbursement policies discourage hospitals from adopting academic algorithms, creating a persistent translational valley.

The findings of this study indicate an increasing trend in publications in this field, with Chinese and American scholars achieving significant accomplishments. Nevertheless, there is generally a lack of extensive collaboration, while deep learning and multimodal fusion remain key research hotspots, consistent with previous findings in similar studies (40). The innovative aspect is we conducted a further in-depth analysis of the development trends in research hotspots and explored the underlying reasons for the mismatch between output quantity and quality, as well as the low clinical translation rate. Overall, regarding research and analysis on the application of AI in CT imaging, most studies focus on a single disease, primarily involving diagnosis of pulmonary and cardiovascular diseases, as well as diagnosis and target region delineation or image segmentation for hepatobiliary diseases (41-44). Only a few articles are not limited to specific diseases and provide a comprehensive analysis of research trends. In similar studies, most researchers utilized the WoSCC database for analysis, which aligns with our database usage (40). Some authors from China additionally incorporated the CNKI database to provide evidence comparing China’s research progress in this field with international developments (45). However, our study employed only a single database, resulting in relatively insufficient evidence for comparisons between China and international standards.

The limitations of this study are listed as follows: (I) the study failed to incorporate the latest 2025 research data, resulting in an inability to promptly reflect the shift in research focus. This dataset gap may underestimate the booming trends of generative medical foundation models and multi-center clinical validation research, while preventing comprehensive verification of the long-term decline of temporary hotspots such as COVID-19 CT imaging. In addition, the absence of high-impact papers and cross-border medical-engineering collaborations published in 2025 weakens the timeliness of comparative analysis between domestic and international research output, and the staged division of field development only reflects the state before 2024 rather than the structural shift toward clinical translation since 2025. (II) Our study can’t reflect the research hotspots regarding specific clinical applications. This study only retrieved literature of AI-assisted CT diagnosis and excluded CT-guided radiotherapy papers, leading to several limitations. The keyword network and hotspot analysis cannot reflect the full landscape of AI-CT applications, omitting radiotherapy research on target segmentation and dose optimization. Besides, the structural differences between domestic and international research layout are insufficiently discussed, and the research gaps specific to radiotherapy are not summarized, which narrows the comprehensive understanding of the whole field. (III) Different databases contain varying article datasets, which may significantly influence analytical outcomes. The WoSCC Core database stands out for its widely recognized authority and rigorous data quality, which serves as the cornerstone of early bibliometric development, making it the primary database employed in this study. However, relying solely on a single database may lead to biased analytical results due to selective inclusion practices. (IV) Another limitation of this bibliometric analysis is that all review papers were excluded from the final dataset. As highly summative pieces in the field, reviews play an irreplaceable role in helping readers understand the development of the field. Reviews tend to accumulate far more citations and integrate massive scattered primary research themes, which would severely interfere with the accurate measurement of publication output, keyword centrality and temporal hotspot shifts of original empirical studies. To guarantee the objectivity and authenticity of our analytical results focusing on original AI-CT research, we filtered out review literature in advance.

Finally, the research only focused on the application of AI in CT and does not represent the overall progress in the field of imaging.

Conclusions

Based on this analysis, we draw the following conclusions: (I) AI is widely applied and will remain a focal point of ongoing interest and research in Medical Imaging research; it is essential to strengthen collaboration among researchers and research institutions, particularly international collaborations, to incorporate advanced research methodologies and thus improve the quality of AI research in Medical Imaging. (II) The leading research in AI applied to medical imaging primarily encompasses radiomics, machine learning, and deep learning. Future research should build upon existing studies and be firmly grounded in clinical practice, aiming to leverage AI to enhance traditional diagnostic methodologies. (III) The study identified gaps in several areas that have not been sufficiently investigated, including algorithm design, clinical translation, data infrastructure, and ethical research. Future efforts could focus on developing interpretable, lightweight, and multimodal fusion intelligent algorithms, conducting multicenter prospective clinical validation, standardizing image annotation and diagnostic criteria, establishing a unified shared CT database, advancing ethical research on imaging data, expanding application scenarios such as multi-organ assessment and postoperative efficacy evaluation, and strengthening medical-engineering collaboration along with health economics evaluation to facilitate the clinical implementation of AI-powered CT diagnostic technologies.

Acknowledgments

We sincerely appreciate the authors’ dedicated efforts in the research process and the strong support from our institution. We are deeply grateful to Zuo Houdong for his guidance and continuous support throughout this research.

Footnote

Reporting Checklist: The authors have completed the BIBLIO reporting checklist. Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2026-1-0015/rc

Peer Review File: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2026-1-0015/prf

Funding: This study was supported by Chengdu Municipal Health Commission (No. 2026425).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2026-1-0015/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. This bibliometric study extracted metadata of published papers from the Web of Science Core Collection (WoSCC). All analyzed data are publicly available academic information, and no private personal data, clinical samples, patient records or animal experimental materials were involved in this research. Since no human or animal subjects were included, ethical committee approval and informed consent were not required in this study.

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. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature 2015;521:436-44. [Crossref] [PubMed]
2. Brady AP. The vanishing radiologist-an unseen danger, and a danger of being unseen. Eur Radiol 2021;31:5998-6000. [Crossref] [PubMed]
3. HeKZhangXRenSDeep residual learning for image recognition. arXiv:1512.03385
4. Fu Y, Lei Y, Wang T, et al. A review of deep learning based methods for medical image multi-organ segmentation. Phys Med 2021;85:107-22. [Crossref] [PubMed]
5. Huang J, Xiang Y, Gan S, et al. Application of artificial intelligence in medical imaging for tumor diagnosis and treatment: a comprehensive approach. Discov Oncol 2025;16:1625. [Crossref] [PubMed]
6. Wang J, Wang K, Yu Y, et al. Self-improving generative foundation model for synthetic medical image generation and clinical applications. Nat Med 2025;31:609-17. [Crossref] [PubMed]
7. Chen Y, Ng L. Advancing Multimodal Medical Image Generation: A Self-Improving Generative Foundation Model. MedComm (2020) 2025;6:e70490.
8. Ninkov A, Frank JR, Maggio LA. Bibliometrics: Methods for studying academic publishing. Perspect Med Educ 2022;11:173-6. [Crossref] [PubMed]
9. Xia GH, Wang Q, Ruan XP, et al. Research trends and focus in low-dose CT study at home and abroad based on Web of Science. Chin J Med Imaging 2013:457-463.
10. Ozturk T, Talo M, Yildirim EA, et al. Automated detection of COVID-19 cases using deep neural networks with X-ray images. Comput Biol Med 2020;121:103792. [Crossref] [PubMed]
11. Li L, Qin L, Xu Z, et al. Using Artificial Intelligence to Detect COVID-19 and Community-acquired Pneumonia Based on Pulmonary CT: Evaluation of the Diagnostic Accuracy. Radiology 2020;296:E65-71. [Crossref] [PubMed]
12. MeiXLeeHCDiaoKArtificial intelligence-enabled rapid diagnosis of COVID-19 patients.medRxiv 2020;2020.04.12.20062661.
13. Zhang K, Liu X, Shen J, et al. Clinically Applicable AI System for Accurate Diagnosis, Quantitative Measurements, and Prognosis of COVID-19 Pneumonia Using Computed Tomography. Cell 2020;181:1423-1433.e11. [Crossref] [PubMed]
14. Oh Y, Park S, Ye JC. Deep Learning COVID-19 Features on CXR Using Limited Training Data Sets. IEEE Trans Med Imaging 2020;39:2688-700. [Crossref] [PubMed]
15. Ardakani AA, Kanafi AR, Acharya UR, et al. Application of deep learning technique to manage COVID-19 in routine clinical practice using CT images: Results of 10 convolutional neural networks. Comput Biol Med 2020;121:103795. [Crossref] [PubMed]
16. Harmon SA, Sanford TH, Xu S, et al. Artificial intelligence for the detection of COVID-19 pneumonia on chest CT using multinational datasets. Nat Commun 2020;11:4080. [Crossref] [PubMed]
17. van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res 2017;77:e104-7. [Crossref] [PubMed]
18. Gupta R, Srivastava D, Sahu M, et al. Artificial intelligence to deep learning: machine intelligence approach for drug discovery. Mol Divers 2021;25:1315-60. [Crossref] [PubMed]
19. Mahmud M, Kaiser MS, Hussain A, et al. Applications of Deep Learning and Reinforcement Learning to Biological Data. IEEE Trans Neural Netw Learn Syst 2018;29:2063-79. [Crossref] [PubMed]
20. Xu Y, Hosny A, Zeleznik R, et al. Deep Learning Predicts Lung Cancer Treatment Response from Serial Medical Imaging. Clin Cancer Res 2019;25:3266-75. [Crossref] [PubMed]
21. Yang G, Ye Q, Xia J. Unbox the black-box for the medical explainable AI via multi-modal and multi-centre data fusion: A mini-review, two showcases and beyond. Inf Fusion 2022;77:29-52. [Crossref] [PubMed]
22. Giger ML. Machine Learning in Medical Imaging. J Am Coll Radiol 2018;15:512-20. [Crossref] [PubMed]
23. 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]
24. Kim DH, MacKinnon T. Artificial intelligence in fracture detection: transfer learning from deep convolutional neural networks. Clin Radiol 2018;73:439-45. [Crossref] [PubMed]
25. Trebeschi S, Drago SG, Birkbak NJ, et al. Predicting response to cancer immunotherapy using noninvasive radiomic biomarkers. Ann Oncol 2019;30:998-1004. [Crossref] [PubMed]
26. Rios Velazquez E, Parmar C, Liu Y, et al. Somatic Mutations Drive Distinct Imaging Phenotypes in Lung Cancer. Cancer Res 2017;77:3922-30. [Crossref] [PubMed]
27. Bian Y, Li J, Ye C, et al. Artificial intelligence in medical imaging: From task-specific models to large-scale foundation models. Chin Med J (Engl) 2025;138:651-63. [Crossref] [PubMed]
28. Aromiwura AA, Settle T, Umer M, et al. Artificial intelligence in cardiac computed tomography. Prog Cardiovasc Dis 2023;81:54-77. [Crossref] [PubMed]
29. Wang S, Cao G, Wang Y, et al. Review and Prospect: Artificial Intelligence in Advanced Medical Imaging. Front Radiol 2021;1:781868. [Crossref] [PubMed]
30. Immonen E, Wong J, Nieminen M, et al. The use of deep learning towards dose optimization in low-dose computed tomography: A scoping review. Radiography (Lond) 2022;28:208-14. [Crossref] [PubMed]
31. Zhang J, Yang M, Guo W, et al. Detection-guided deep learning-based model with spatial regularization for lung nodule segmentation. Quant Imaging Med Surg 2025;15:4204-16. [Crossref] [PubMed]
32. Giarletta L, Zhou B, Marano R, et al. Artificial Intelligence in Coronary Computed Tomography: Current Applications, Future Potentials, and Real-world Challenges. J Thorac Imaging 2026;41:e0873. [Crossref] [PubMed]
33. Liu J, Fan W, Yang Y, et al. Deep learning-based identification and localization of intracranial hemorrhage in patients using a large annotated head computed tomography dataset: a retrospective multicenter study. Intell Med 2025;5:14-22.
34. Costantini P, Groenhoff L, Ostillio E, et al. Advancements in Cardiac CT Imaging: The Era of Artificial Intelligence. Echocardiography 2024;41:e70042. [Crossref] [PubMed]
35. Feng G, Zhao Y, Yan F, et al. Escaping the data misuse maze: reorienting medical research toward clinical needs. Innov Med 2025;3:100153.
36. Chen F, Yang X, Yin K, et al. Innovative strategies for interdisciplinary medical-engineering education in China. Glob Med Educ 2025;2:213-226.
37. Krishnaswamy D, Bontempi D, Thiriveedhi VK, et al. Enrichment of lung cancer computed tomography collections with AI-derived annotations. Sci Data 2024;11:25. [Crossref] [PubMed]
38. Chinese Society of Radiology Medical Imaging AI Working Group. Expert consensus on construction and annotation of multi-center CT database for advanced gastric cancer. Chin J Radiol 2025;59:861-871.
39. Liao X, Yao C, Jin F, et al. Barriers and facilitators to implementing imaging-based diagnostic artificial intelligence-assisted decision-making software in hospitals in China: a qualitative study using the updated Consolidated Framework for Implementation Research. BMJ Open 2024;14:e084398. [Crossref] [PubMed]
40. Zhu K, Shen Z, Wang M, et al. Visual Knowledge Domain of Artificial Intelligence in Computed Tomography: A Review Based on Bibliometric Analysis. J Comput Assist Tomogr 2024;48:652-62. [Crossref] [PubMed]
41. Xia Y, Zhong S, Lou X, et al. Research trends of artificial intelligence and radiomics in lung cancer: a bibliometric analysis. Quant Imaging Med Surg 2024;14:8771-84. [Crossref] [PubMed]
42. Li M, Sun S, Liu M, et al. Mapping knowledge landscapes and emerging trends in AI for coronary artery disease imaging biomarkers: A bibliometric and visualization analysis. Curr Probl Cardiol 2026;51:103302. [Crossref] [PubMed]
43. Zheng RJ, Li DL, Lin HM, et al. Bibliometrics of artificial intelligence applications in hepatobiliary surgery from 2014 to 2024. World J Gastrointest Surg 2025;17:104728. [Crossref] [PubMed]
44. Delmoral JC. R S Tavares JM. Semantic Segmentation of CT Liver Structures: A Systematic Review of Recent Trends and Bibliometric Analysis : Neural Network-based Methods for Liver Semantic Segmentation. J Med Syst 2024;48:97. [Crossref] [PubMed]
45. Zhao X, Zhang F. Visual analysis of AI research hotspots in the field of medical imaging. China Digit Med 2023;18:83-89.