What this is
- This research analyzes the trends and characteristics of artificial intelligence (AI) applications in oncology from 1992 to 2022.
- It employs bibliometric methods to evaluate the growth and hotspots in AI-related oncology literature.
- The findings reveal significant increases in publications and highlight key authors, journals, and research themes.
Essence
- AI applications in oncology have seen substantial growth, particularly after 2015, with a focus on machine learning and radiomics. Key contributors include leading institutions and prolific authors, underscoring the field's evolving landscape.
Key takeaways
- The number of AI-related oncology publications surged significantly after 2015, indicating growing interest and investment in this area.
- The most prolific authors include Esteva A and Gillies RJ, with their work contributing to a substantial portion of total citations.
- Research hotspots identified include machine learning, deep learning, and radiomics, which are pivotal in advancing cancer diagnosis and treatment.
Caveats
- The analysis relies solely on data from the Web of Science, which may not represent the complete landscape of AI publications in oncology.
- The study's focus on English-language publications may overlook significant contributions from non-English sources.
Definitions
- Bibliometric analysis: A method that quantitatively analyzes academic literature to assess research trends and impacts.
AI simplified
Introduction
Artificial intelligence is based on creating machines that think like humans and is seen as the apotheosis of science [1]. Artificial intelligence (AI), which has emerged in different fields in the twenty-first century, has become a trend in many fields, such as medicine, science, and business [2].
The integration of artificial intelligence (AI) into cancer treatment has also gained momentum, with AI-powered predictive models showing promise in identifying patients who are most likely to respond to specific therapies. Recent studies have demonstrated the potential of AI to predict response to checkpoint inhibitors in non-small cell lung cancer [3], predict response to chemotherapy in breast cancer [4], identify high-risk patients who may benefit from early intervention [5], develop personalized treatment plans for patients with rare cancers such as pancreatic cancer [6], and predict treatment outcomes in patients with melanoma [7]. As the field of cancer treatment continues to evolve, it is essential that we stay up-to-date with the latest advancements and explore the potential applications of AI in predicting response to cancer therapies.
It is possible to categorize medical data that can be assessed with deep learning/machine learning/artificial intelligence into genetic, imaging, and clinical data. The images may be real-size photographs of visible parts of the human body and internal parts of the human body obtained by devices such as computed tomography, x-ray, magnetic resonance, and endoscopic devices. Radiomics is a research field that applies advanced image analysis techniques to extract quantitative features from medical imaging data.. In 2009, “Radiogenomics: Creating a link between Molecular Diagnostics and Diagnostic Imaging” [8] and “Radiomics: Images Are More than Pictures, They Are Data” [9] in, the term radiomics started to appear in the literature.
Competitions using The Pascal Visual Object Classes Challenge and Imagenet datasets have been held worldwide to recognize highly accurate images by electronic systems. In 2012, AlexNet software, which participated in the competition under the name Supervision, won this competition by a large margin, and the paper "ImageNet classification with deep convolutional neural networks," published by the authors [10], was an important milestone. Please note that, Ilya Sutskever is a major contributor to ChatGPT, a large language model based AI software. The article "Radiomics: images are more than pictures, they are data” [9], drew attention to the fact that meaningful results can be obtained primarily in cancer patients by converting radiologic images into numerical data. Publications on the ability of Alexnet-type software to successfully evaluate patient-related images have started to appear since 2017 and have increased steadily.
Magnified images of cells in the human body tissues can be obtained with devices such as endoscopy, or cells extracted from the body by biopsy can be examined by various processes and staining. The field that deals with microscopic images of tissues are called pathomics [11]. There are very few publications using the term pathomics.
Genetic data can be related to DNA gene sequences or protein structures. The field dealing with gene sequences is called genomics and the field dealing with protein structures is called proteomics. The highly cited publications on gene sequencing and cancer started even before radiomics and received many citations [12]. Genetic data can be about DNA gene sequences or protein structures. The field that deals with gene sequences is called genomics, and the field that deals with protein structures is called proteomics. Highly cited publications on gene sequencing and cancer started even before radiomics [12, 13]. Although not cited as genomics, publications on proteomics started appearing before radiomics [14].
While the name clinicomics seems appropriate for evaluating clinical data with machine learning [15–17], the term has only recently been used. In the USA., clinical data obtained from hospitals' electronic health records (EHR) are anonymized/de-identified in a CancerLinq database and re-recorded so that meaningful conclusions can be drawn and used by users who want to benefit from it [18, 19]. Unlike the SEER (Surveillance, Epidemiology, and End Results) database (https://seer.cancer.gov↗, an extensive anonymized cancer patient database), where fewer parameters about a patient are stored, it should be considered to allow for the evaluation of individualized diagnoses and treatments. Clinicomics may attempt to extract meaningful data from text based clinical records and is expected to utilize large language model AI frameworks.
Since it is necessary to use one of the research methods appropriate for evaluation and prediction, bibliometric analysis is one of the analyses that serves this purpose. Bibliometrics is used as a type of analysis that systematically analyzes datasets [20]. This type of analysis requires a systematic literature review and meaningful structuring of a large dataset [20]. Bibliometric analysis is a method used to analyze the studies conducted in a specific field, and its use has increased in recent years.
While bibliometric analysis in this concept is seen as inevitable for many fields [21], it is necessary to examine academic studies on the impact of artificial intelligence applications in medicine through bibliometric analysis. This is because AI's increasing popularity and widespread use cause the subject to become a need in terms of health. This study analyzed 7923 articles published between 1992 and 2022 with the R program, biblioshiny package program, and VOSviewer software [22]. The article contributed to determining the techniques used in the cumulative information in the relevant research, thematic analysis, and central trends. It sheds light on the research to be conducted. In this context, the study will first include the literature on artificial intelligence and cancer, then the data used and the research method will be mentioned.
Karger [23] provides a bibliometric analysis of research on AI for cancer detection, emphasizing its growth and potential for early detection. Khanam and Kumar [24] explore the recent applications of AI algorithms, including machine learning and deep learning, in the early detection of various cancers. Pacurari et al. [25] focus on using AI techniques, such as support vector machines and neural networks, to detect and classify lung, breast, and brain cancers using medical imaging. Overall, these papers highlight the importance of AI in improving Oncology and emphasize the need for further research in this field.
The objective of this research is to address specific inquiries regarding the utilization of Artificial Intelligence (AI) in Oncology. Sources: Which scientific journals are most influential in the field? Researhers: Which Authors are the most influential in the field of Artificial Intelligence (AI) applications in Oncology? Papers: Which papers about Artificial Intelligence (AI) applications in Oncology? Keywords: What are the most popular authors keywords in Artificial Intelligence (AI) applications in Oncology research? How have the themes of Artificial Intelligence (AI) applications in Oncology evolved? Funding: Which fundings are the most influential in the field of Artificial Intelligence (AI) applications in Oncology?
Method
The relevant subject is examined statistically and mathematically in Bibliometrics, and a framework is drawn for the course. In addition, bibliometric analysis can also reveal the effectiveness of studies conducted through statistical data [26]. The data for this study were retrieved from the online database Web of Science on 21.11.2023. The reasons for using the Web of Science database instead of Scopus Google Scholar for bibliometric analysis are that it is the most extensive database in abstracts and literature, it can produce information with better actions, decisions, and results [27], it is a valuable resource for bibliometric studies, and it offers a comprehensive perspective on publications in the fields of science, technology, art, medicine, and social sciences. The Web of Science database used in bibliometric analysis is preferred [28].
The search strategy of this study is as follows: While obtaining the data of the study, the first step was to search the Web of Science database (TI = ("deep-Learn*" OR "machine learn*" OR "deep learn*"OR "artificial intelligence" OR "artificial neural network*" OR "deep neural network*" OR radiomics OR pathomics)) AND (AB = (melanoma OR cancer OR malignancy OR leukemia OR lymphoma OR neoplasia OR SEER* OR "Surveillance, Epidemiology, and End Results" OR CancerLinQ*)) AND (PY = 1992–2022) AND (DT = Article).
Data from databases were directly accessed to retrieve metadata from chosen documents, encompassing details like active authors, journal sources, countries, institutions, and funding sources. The analysis involved employing tools such as Bibliometrix[29] package (version 3.1.4, http://www.bibliometrix.org↗) in R software (version 3.6.3), VOSviewer [22], and Litmaps [30] to generate comprehensive bibliometric insights. Initially, raw data in plain text format was loaded and processed, involving calculations and visual representations of metadata comprising sources, authors, and citations. This was followed by intricate analyses focusing on clustering and the conceptual framework encompassing intelligence and society. Additionally, a comprehensive bibliometric evaluation was carried out, encompassing co-authorship and keyword co-occurrence using the VOSviewer software [22].
Results
When 7923 articles were reviewed, although the importance of genetic data in cancer diagnosis is significant, there are relatively few publications in machine learning and genomics fields, comprising only 2.2% (174 articles) of the studies in this work. There are only 12 articles related with pathomics, and deep learning in WOS database. The term 'Clinicomics' has not yet been established and has not found a place within the article set of this study. When searching for the term 'Clinical informatics,' only four articles were found; 'Clinical data extraction' yielded 1.6% (124 articles), 'SEER' resulted in 0.90% (71 articles), and 'NLP' led to 28 articles. Only 1 article containing the term 'CancerLing' [31] was found. Only three articles can be found when the keywords 'CancerLinq' and 'machine learning' are searched across all fields in the WOS database. As 1716 articles were found with the term 'Radiomics' and 2215 articles with the term 'image,' we can conclude that the most commonly used datasets related to the topic of this article are primarily in the form of images.
General information
The findings of the bibliometric analysis conducted within the scope of the study will be presented in this section.
The Web of Science was used to meticulously define and classify citation topics. All findings from the search query were included in the review without further filtration. The citation topics were narrowly focused and categorized according to the recently published classifications by the Web of Science, encompassing over 2500 detailed citation topics. This classification operates hierarchically below the Web of Science subject categories and citation topics at a broader level, enabling a precise and unbiased evaluation of the technologies utilized in the search query.
Distribution of WOS articles and Citation on Artificial Intelligence (AI) and Oncology between 1992 and 2022 (n = 7923) (InCites, 2023)
Distribution of documents in Q1, Q2, Q3, and Q4 Journals on Artificial Intelligence (AI) and Oncology between 1992 and 2022 (InCites, 2023)
The most common types of diseases mentioned in the articles were Artificial Intelligence and Oncology (Web of Science, 2023)
| Description | Results |
|---|---|
| Main Information About Data | |
| Timespan | 1992:2022 |
| Sources (Journals, Books, etc.) | 1592 |
| Documents | 7923 |
| Annual Growth Rate % | 19.34 |
| Average citations per doc | 24 |
| Document Contents | |
| Keywords Plus (ID) | 7443 |
| Author's Keywords (DE) | 10,966 |
| Authors | |
| Authors | 35,198 |
| Authors of single-authored docs | 120 |
| Co-Authors per Doc | 7.95 |
| International co-authorships % | 29.48 |
Authors
Figure 5b shows the main statistical characteristics of the Top 20 authors ranked by number of citations. When the graph is analyzed, Aerts Hugo and Gillies Robyn are the leading authors working on artificial intelligence and oncology. The author's number of citations covers 22.7 percent of the total citations.
Lotka's Law and Author Productivity Rate (Bibliometrix & R software, 2023)
Top 20 Authors published the most AI articles in oncology from 1992 to 2022.Top 20 Authors published the most AI citations in oncology from 1992 to 2022 (InCites, 2023) a b
Historiography of AI articles in oncology from 1992 to 2022 (Bibliometrix & R software, 2023)
| Documents written | No. of authors | Proportion of Authors |
|---|---|---|
| 1 | 25,587 | 0.729 |
| 2 | 4999 | 0.142 |
| 3 | 1799 | 0.051 |
| 4 | 921 | 0.026 |
| 5 | 553 | 0.016 |
| 6 | 313 | 0.009 |
| 7 | 222 | 0.006 |
| 8 | 136 | 0.004 |
| 9 | 102 | 0.003 |
| 10 | 99 | 0.003 |
Papers
For this study, Litmaps, an advanced science discovery platform known for its visual citation navigation, has been utilized. This platform offers an interface facilitating the exploration of scientific literature, enabling researchers to delve into the research terrain and uncover articles intricately linked within maps. Litmaps also presents convenient options for swiftly importing articles through various means such as reference manager, keyword search, ORCID ID, DOI, or by utilizing a seed article [36].
![Seed Maps of Esteva [].Seed Maps of Gillies [] (Litmaps, 2023) a b [37] [9]](https://europepmc.org/articles/PMC11480271/bin/12672_2024_1415_Fig7_HTML.jpg.jpg "Click to view full size")
Seed Maps of Esteva [].Seed Maps of Gillies [] (Litmaps, 2023) a b [37] [9]
| Paper | Article Title | No. of citation | DOI | Journal | WoS categories |
|---|---|---|---|---|---|
| Esteva et al., 2017 | Dermatologist-level classification of skin cancer with deep neural networks | 5900 | 10.1038/nature21056 | Nature | Multidisciplinary Sciences |
| Gillies et al., 2016 | Radiomics: Images Are More than Pictures, They Are Data | 4362 | 10.1148/radiol.2015151169 | Radiology | Radiology |
| Lambin et al., 2012 | Radiomics: Extracting more information from medical images using advanced feature analysis | 3018 | 10.1016/j.ejca.2011.11.036 | Eur. J. Cancer | Oncology |
| Aerts et al., 2014 | Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach | 2958 | 10.1038/ncomms5006 | Nat. Commun | Multidisciplinary Sciences |
| Van Griethuysen et al., 2017 | Computational Radiomics System to Decode the Radiographic Phenotype | 2725 | 10.1158/0008-5472.CAN-17-0339 | Cancer Res | Oncology |
| Khan et al., 2001 | Classification and diagnostic prediction of cancers using gene expression profiling and artificial neural networks | 1897 | 10.1038/89044 | Nat. Med | Biochemistry & Molecular Biology |
| Shipp et al., 2002 | Diffuse large B-cell lymphoma outcome prediction by gene-expression profiling and supervised machine learning | 1796 | 10.1038/nm0102-68 | Nat. Med | Biochemistry & Molecular Biology |
| Bejnordi et al., 2017 | Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer | 1466 | 10.1001/jama.2017.14585 | JAMA-J. Am. Med. Assoc | Medicine, General & Internal |
| Kumar et al.,2012 | Radiomics: the process and the challenges | 1374 | 10.1016/j.mri.2012.06.010 | Magn. Reson. Imaging | Radiology |
| Zwanenburg et al., 2020 | The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping | 1367 | 10.1148/radiol.2020191145 | Radiology | Radiology |
| Coudray et al., 2018 | Classification and mutation prediction from non-small cell lung cancer histopathology images using deep learning | 1243 | 10.1038/s41591-018-0177-5 | Nat. Med | Biochemistry & Molecular Biology |
| Huang et al., 2016 | Development and Validation of a Radiomics Nomogram for Preoperative Prediction of Lymph Node Metastasis in Colorectal Cancer | 1151 | 10.1200/JCO.2015.65.9128 | J. Clin. Oncol | Oncology |
| Malta et al., 2018 | Machine Learning Identifies Stemness Features Associated with Oncogenic Dedifferentiation | 1009 | 10.1016/j.cell.2018.03.034 | Cell | Biochemistry & Molecular Biology |
| Campanella et al., 2019 | Clinical-grade computational pathology using weakly supervised deep learning on whole slide images | 899 | 10.1038/s41591-019-0508-1 | Nat. Med | Biochemistry & Molecular Biology |
| Johnson et al., 2019 | Survey on deep learning with class imbalance | 892 | 10.1186/s40537-019-0192-5 | J. Big Data | Computer Science |
| Ardila et al., 2019 | End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography | 828 | 10.1038/s41591-019-0447-x | Nat. Med | Biochemistry & Molecular Biology |
| Sirinukunwattana et al., 2016 | Locality Sensitive Deep Learning for Detection and Classification of Nuclei in Routine Colon Cancer Histology Images | 706 | 10.1109/TMI.2016.2525803 | IEEE Trans. Med. Imaging | Engineering, Biomedical |
| Ye et al., 2013 | Predicting hepatitis B virus-positive metastatic hepatocellular carcinomas using gene expression profiling and supervised machine learning | 699 | 10.1038/nm843 | Nat. Med | Biochemistry & Molecular Biology |
| Bi et al., 2019 | Artificial intelligence in cancer imaging: Clinical challenges and applications | 698 | 10.3322/caac.21552 | CA-Cancer J. Clin | Oncology |
| Sun et al., 2018 | A radiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study | 638 | 10.1016/S1470-2045(18)30,413-3 | Lancet Oncol | Oncology |
| Parmar et al., 2015 | Machine Learning methods for Quantitative Radiomic Biomarkers | 610 | 10.1038/srep13087 | Sci Rep | Multidisciplinary Sciences |
| Yao & Liu, 1997 | A new evolutionary system for evolving artificial neural networks | 591 | 10.1109/72.572107 | IEEE Trans. Neural Netw | Artificial Intelligence |
| Litjens et al., 216 | Deep learning as a tool for increased accuracy and efficiency of histopathological diagnosis | 588 | 10.1038/srep26286 | Sci Rep | Multidisciplinary Sciences |
| Rajpurka et al., 2018 | Deep learning for chest radiograph diagnosis: A retrospective comparison of the CheXNeXt algorithm to practicing radiologists | 545 | 10.1371/journal.pmed.1002686 | PLos Med | Medicine, General & Internal |
| Bera et al., 2019 | Artificial intelligence in digital pathology—new tools for diagnosis and precision oncology | 525 | 10.1038/s41571-019-0252-y | Nat. Rev. Clin. Oncol | Oncology |
Sources (Journals)
In Fig. 7, Bradford's Law divides journals into three main classes and helps to find the core journals. First formulated in 1934, Bradford's Scatter Law "describes the scatter or distribution of literature on a particular topic across journals" [34]. According to this law, there should be an inverse relationship between the number of studies published on a topic and the number of journals in which they are published. Journals are divided into regions by ranking them according to the number of studies they publish. Although the number of journals in each region is not equal, the total number of publications in the regions will be equal. Because the productivity of journals is different from each other [38]. As a result of the analysis, it was determined that the journals were divided into three regions. Figure 7 shows the sources in the first region.
When we look at the change in scientific journals over the years, it was first published in Scientific Reports in 2015. After 2020, a significant increase in publications in Artificial Intelligence (AI) applications in Oncology was observed. Frontiers in Oncology journal is leading this increase.
As shown in Fig. 8b, Scientific Journals began publishing artificial intelligence studies in the field of Oncology in 2015. It began to rise after 2018. In the early part of 2020, Scientigfic Reports was leading in publishing articles in this area, but after 2020, Frontiers in Oncology took the lead. Figure 15 shows that Frontiers in Oncology published these studies at significantly higher rates compared to other Journals.
Distribution of the scientific journals publishing the most articles in artificial intelligence and Oncology according to Bradford's law.Distribution of the journals with the most articles by year (Bibliometrix & R software, 2023) a b
Web of science categories
| Name | Web of Science Documents | Times Cited | Citation Impact | Documents in Q1 Journals | Documents in Q2 Journals | Documents in Q3 Journals | Documents in Q4 Journals |
|---|---|---|---|---|---|---|---|
| ONCOLOGY | 2021 | 49,877 | 0.04 | 538 | 1006 | 195 | 114 |
| RADIOLOGY, NUCLEAR MEDICINE & MEDICAL IMAGING | 1774 | 53,085 | 0.03 | 763 | 550 | 239 | 51 |
| ENGINEERING, BIOMEDICAL | 493 | 14,545 | 0.03 | 227 | 151 | 59 | 21 |
| COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE | 490 | 19,532 | 0.03 | 159 | 136 | 58 | 33 |
| MATHEMATICAL & COMPUTATIONAL BIOLOGY | 477 | 8409 | 0.06 | 242 | 82 | 22 | 32 |
| COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS | 459 | 14,204 | 0.03 | 276 | 83 | 39 | 10 |
| ENGINEERING, ELECTRICAL & ELECTRONIC | 440 | 10,731 | 0.04 | 98 | 232 | 77 | 15 |
| COMPUTER SCIENCE, INFORMATION SYSTEMS | 418 | 6664 | 0.06 | 91 | 137 | 105 | 20 |
| MEDICAL INFORMATICS | 322 | 8318 | 0.04 | 147 | 67 | 59 | 11 |
| BIOCHEMICAL RESEARCH METHODS | 307 | 6115 | 0.05 | 125 | 111 | 40 | 18 |
| BIOTECHNOLOGY & APPLIED MICROBIOLOGY | 268 | 5080 | 0.05 | 100 | 59 | 61 | 7 |
| MEDICINE, RESEARCH & EXPERIMENTAL | 246 | 12,342 | 0.02 | 88 | 38 | 44 | 28 |
| BIOCHEMISTRY & MOLECULAR BIOLOGY | 241 | 12,785 | 0.02 | 119 | 75 | 31 | 9 |
| COMPUTER SCIENCE, THEORY & METHODS | 236 | 5488 | 0.04 | 65 | 90 | 20 | 3 |
| HEALTH CARE SCIENCES & SERVICES | 225 | 3944 | 0.06 | 76 | 75 | 31 | 11 |
| GASTROENTEROLOGY & HEPATOLOGY | 217 | 5324 | 0.04 | 84 | 60 | 31 | 21 |
| CHEMISTRY, MULTIDISCIPLINARY | 201 | 2710 | 0.07 | 27 | 82 | 86 | 2 |
| SURGERY | 193 | 2553 | 0.08 | 90 | 61 | 20 | 9 |
| GENETICS & HEREDITY | 165 | 2500 | 0.07 | 54 | 78 | 20 | 4 |
| BIOLOGY | 163 | 3630 | 0.04 | 101 | 38 | 16 | 3 |
Keywords
Figure 10 shows the visualization network map of author keywords co-occurrence. Four Clusters are formed weights based on occurrences. The red color indicates Cluster 1 (radiomics, magnetic resonance imaging, computer tomograph, etc.); the green color indicates Cluster 2 (machine learning, prediction, artificial neural network, etc.); the blue color indicates Cluster 3 (deep learning, breast cancer, artificial intelligence, etc.); the yellow color indicates Cluster 4 (cancer, feature extraction, image classification, etc.).
Most Frequent Author Keywords in the AI and Oncology literature, 1992–2022 (Bibliometrix & R software, 2023)
Network map of author keywords co-occurrence that appeared 12+ times in the artificial intelligence and Oncology literature, 1992–2022 (VOSviever, 2023)
Countries and universities
Figure 14 further illustrates that the majority of publications are authored by individuals from the same countries. This trend might arise from authors' preferences to collaborate within their research groups or with academics sharing the same national background.
Network of institutions among articles with authors affiliated with institutions (Bibliometrix & R software, 2023)
Distribution of author affiliations to organizations based on the selected dataset (top 20 organizations) (Bibliometrix & R software, 2023)
Co-authorship country visualization network map (Wosviever, 2023)
Most productive countries: Single Country Publications (SCP), Multiple Country Publications (MCP) (Bibliometrix & R software, 2023)
| Country | Articles | SCP | MCP | Freq | MCP_Ratio |
|---|---|---|---|---|---|
| CHINA | 2508 | 2071 | 437 | 0.323 | 0.174 |
| USA | 1428 | 994 | 434 | 0.184 | 0.304 |
| INDIA | 461 | 383 | 78 | 0.059 | 0.169 |
| KOREA | 422 | 343 | 79 | 0.054 | 0.187 |
| JAPAN | 286 | 251 | 35 | 0.037 | 0.122 |
| GERMANY | 263 | 145 | 118 | 0.034 | 0.449 |
| ITALY | 253 | 187 | 66 | 0.033 | 0.261 |
| UNITED KINGDOM | 222 | 108 | 114 | 0.029 | 0.514 |
| CANADA | 180 | 117 | 63 | 0.023 | 0.35 |
| NETHERLANDS | 155 | 70 | 85 | 0.02 | 0.548 |
| FRANCE | 130 | 81 | 49 | 0.017 | 0.377 |
| IRAN | 104 | 69 | 35 | 0.013 | 0.337 |
| TURKEY | 96 | 85 | 11 | 0.012 | 0.115 |
| AUSTRALIA | 95 | 40 | 55 | 0.012 | 0.579 |
| SAUDI ARABIA | 85 | 41 | 44 | 0.011 | 0.518 |
| EGYPT | 79 | 42 | 37 | 0.01 | 0.468 |
| SPAIN | 77 | 52 | 25 | 0.01 | 0.325 |
| PAKISTAN | 61 | 13 | 48 | 0.008 | 0.787 |
| SWEDEN | 56 | 31 | 25 | 0.007 | 0.446 |
| SWITZERLAND | 48 | 20 | 28 | 0.006 | 0.583 |
Funding
Changes in the distribution of the articles according to the institutions supporting the articles by years (InCites, 2023)
| Funding | Web of Science Documents | Times Cited | International Collaborations | Domestic Collaborations | Documents in Q1 Journals | Documents in Q2 Journals | Documents in Q3 Journals | Documents in Q4 Journals |
|---|---|---|---|---|---|---|---|---|
| National Natural Science Foundation of China | 1045 | 18,886 | 229 | 626 | 472 | 417 | 109 | 22 |
| Department of Health & Human Services-USA | 656 | 22,221 | 251 | 264 | 353 | 196 | 34 | 6 |
| National Institutes of Health (NIH)-USA | 649 | 21,919 | 248 | 263 | 350 | 195 | 33 | 6 |
| NIH National Cancer Institute (NCI)-USA | 270 | 9855 | 97 | 118 | 141 | 81 | 15 | 2 |
| National Research Foundation of Korea | 205 | 2843 | 35 | 130 | 96 | 83 | 8 | 12 |
| National Science Foundation-USA | 102 | 2549 | 43 | 35 | 56 | 24 | 6 | 1 |
| Ministry of Education Culture Sports Science and Technology-Japan | 89 | 1155 | 29 | 45 | 31 | 39 | 9 | 2 |
| Japan Society for the Promotion of Science | 87 | 1121 | 29 | 43 | 31 | 38 | 9 | 2 |
| Ministry of Science ICT & Future Planning-Republic of Korea | 86 | 1155 | 15 | 56 | 40 | 37 | 3 | 3 |
| European Union-EU | 83 | 1964 | 56 | 18 | 43 | 29 | 3 | 1 |
| Fundamental Research Funds for the Central Universities-China | 81 | 1789 | 25 | 41 | 36 | 34 | 10 | 0 |
| National Natural Science Foundation of Guangdong Province-China | 77 | 1665 | 11 | 56 | 36 | 30 | 6 | 2 |
| Grants-in-Aid for Scientific Research-Japan | 72 | 895 | 20 | 38 | 24 | 33 | 8 | 1 |
| UK Research & Innovation-UK | 71 | 3290 | 41 | 24 | 45 | 18 | 0 | 1 |
| Beijing Natural Science Foundation-China | 68 | 1818 | 16 | 46 | 32 | 31 | 5 | 0 |
| Spanish Government | 66 | 3096 | 29 | 25 | 42 | 18 | 2 | 0 |
| Ministry of Science and Technology-Taiwan | 64 | 960 | 12 | 49 | 34 | 23 | 6 | 0 |
| German Research Foundation (DFG)-Germany | 61 | 2074 | 35 | 23 | 40 | 15 | 3 | 1 |
| Ministry of Science & ICT-Republic of Korea | 61 | 733 | 11 | 35 | 26 | 27 | 4 | 1 |
| China Postdoctoral Science Foundation | 60 | 674 | 16 | 38 | 34 | 24 | 1 | 0 |
| Natural Sciences and Engineering Research Council of Canada | 56 | 1602 | 20 | 25 | 26 | 21 | 4 | 0 |
| European Research Council | 45 | 3609 | 31 | 8 | 30 | 13 | 0 | 0 |
| Medical Research Council UK | 44 | 2902 | 25 | 17 | 29 | 11 | 0 | 1 |
| Natural Science Foundation of Zhejiang Province-China | 42 | 713 | 18 | 21 | 20 | 19 | 3 | 0 |
| Canadian Institutes of Health Research | 40 | 1379 | 11 | 24 | 21 | 13 | 0 | 0 |
Discussion
In recent years, artificial intelligence (AI) has swiftly become an integral part of the medical field, particularly in cancer detection. This study utilized the bibliometrix package of R software and Litmaps visualization software to conduct a thorough bibliometric analysis of AI applications in oncology research over the past 30 years. The objective was to provide a comprehensive understanding of the field. Our analysis objectively and systematically outlined the current status of AI applications, identified developmental trends, and highlighted potential research focal points in cancer detection. This endeavor facilitates scholars in rapidly comprehending the research landscape and offers valuable insights for selecting research topics. The initial phase of the study examined publication trends, covering aspects such as countries, institutions, authors, and journals. Subsequently, cluster analysis was applied to keywords to identify research hotspots within the field.
Based on the analysis of publication trends, there has been a significant surge in the number of publications on AI in Oncology over the past four years. China and the United States emerged as the leading nations regarding the volume of publications in this field. Citation counts, widely recognized as an indicator of professional acknowledgment in scientific work, were extensively used to assess research quality. The United States stood out in terms of both citation counts and international collaborations, with a considerable lead over other countries. Additionally, the university contributing the most publications and citations was based in China, underscoring China's pivotal role and global leadership in this domain. Despite China's considerable volume of publications, the relatively low citation counts suggest a need to enhance the quality and impact of its research [40]. This could be attributed, in part, to the later initiation of AI in Oncology research in China, resulting in comparatively lower international academic influence. Notably, Jie Tian and Zaiyi Liu from China emerged as the most published authors, contributing to 10.3% of the publications. Regarding citations, Hugo Aerts from Stanford University and Robyn Gillies from the University of Washington in the United States received the highest recognition. These findings underscore the importance of quantity, quality, and global impact in advancing AI research in cancer detection.
This is the newest bibliometric study that provides detailed information about published literature on the AI in Oncology. The most active institutions were the Chinese Academy of Science and Harward University, and the most productive countries were China and the USA. The most frequently co-occurrence author keywords were radiomics, machine learning, artificial intelligence, and breast cancer.
This bibliometric study holds potential by offering a comprehensive overview of AI in oncology, identifying research hotspots like radiomics and deep learning, and highlighting future research directions such as extracting meaning from genomic/proteomic/clinicomic data. Significant knowledge gaps exist in effectively utilizing these diverse data types and integrating AI seamlessly into clinical workflows, alongside addressing ethical considerations. Researchers are tackling these challenges through advanced algorithms, improved data standardization, and collaborative efforts to develop user-friendly tools. Over the next five years, the field will likely see increased focus on clinicomics and multi-omics, advancements in early cancer detection and prevention, personalized treatment planning, AI-powered drug discovery, and wider adoption of AI tools in clinical settings, fostered by greater collaboration and standardization across the field.
The outcomes of this study hold value for researchers, policymakers, and educational purposes. Additionally, they offer assistance to funding agencies in evaluating current research trajectories and anticipating future trends in AI within Oncology. Effective AI development and treatment therapy is still a hot zone for future research directions. Three leading journals publish the most articles on artificial intelligence and Oncology: Frontiers in Oncology (428 articles) and Scientific Reports and Cancer (284 and 247 articles, respectively). While these three leading journals accounted for 12% of total articles, the remaining list is well-distributed. The majority, almost 70%, of the articles fit into ten significant categories: Oncology (22%), Radiology Nuclear Medicine Medical Imaging (13.4%), Engineering Biomedical (5.1%), Mathematical Computational Biology (5%),
Limitations
This study has some limitations that should be acknowledged. One of the main limitations is the reliance on WoS and InCites as the sole data sources. While these databases are widely wellknown as authoritative sources of citation data, they may not capture the full range of publications and citations in the field. The exclusion of other data sources, such as Scopus, Pubmed, Google Scholar, or non-English language publications, may have introduced biases into our analysis.
Conclusion
Articles about extracting meaning from radiological/microscopic/real patient images in cancer patients using artificial intelligence have been produced for approximately 6 years and have reached a certain maturity. It is understood that some obstacles are related to deriving meaning from genomic/genomic/proteomic data and doctor notes written in text. As new findings emerge, the association of diseases and treatments using existing classification systems with genomic/proteomic data should be expected to increase geometrically/exponentially. In order to derive meaning from clinical data, it is necessary to create new databases of the NoSQL type for transferring values from biochemistry, tumor markers, drug doses, as well as names of drugs/materials/devices, and text notes written manually with pen or keyboard into artificial intelligence software. Therefore, it is expected that big data derived from electronic health records should be reprocessed and re-archived by developing new standards. It is understood that the CancerLinq database has not contributed to machine learning-related publications so far and may not be able to do so in its current state. It is observed that despite fewer available parameters for individuals in the previously established SEER database, more publications related to machine learning have been made. This situation may stem from a structural difference between the CancerLinq and SEER databases.
The exploration of artificial intelligence in Oncology is in its early phases but is anticipated to progress rapidly. Researchers are currently investigating AI applications in various aspects of Oncology within the medical field, including medicine, diagnosis, therapy, and risk assessment. Implementing artificial intelligence proves effective in mitigating human errors and enhancing work efficiency. This bibliometric study offers a comprehensive overview of AI in Oncology research, focusing on the discipline's current state. This perspective assists researchers in identifying critical areas of interest, cutting-edge developments, and emerging research directions within the field.