Abstract
Objective
This analysis aims to examine studies on artificial intelligence (AI) applications in breast cancer diagnosis through bibliometric methods, focusing on temporal and geographical trends. It contributes to shaping the field's roadmap and helping researchers adapt to technological innovations.
Method
A comprehensive search was conducted in the Web of Science (WOS) database. Bibliometric analyses of data from 2013–2024 were performed using VOSviewer and Bibliometrix R programs.
Results
The analysis included 1537 articles. A significant rise in research activity was observed in 2019. The thematic analysis highlighted topics like histopathology, feature selection, deep learning, and machine learning. India was the most productive country with 405 studies. Keyword analysis showed increased usage of terms like transfer learning, CNN, and radiomics. U.S. was the most cited country with 7511 citations. Concept co-occurrence analysis revealed strong associations between terms such as feature selection, datasets, algorithm performance, and classification methods. Bejnordi's 2017 study was identified as the most influential, with 1909 citations.
Discussion and Conclusion
This study identifies key authors, influential works, and trending topics, offering a broad understanding of the field's structure and evolution. It helps outline the advancements and emerging directions in AI applications for breast cancer diagnosis.
Keywords
Introduction
The concept of artificial intelligence (AI) was first defined by Alan Turing in 1950. AI is a structure created by humans that is called intelligent and encompasses behaviors performed by a machine. 1 In the last decade, AI has significantly contributed to solving various biomedical problems, including cancer. 2 AI is a powerful tool that can help accelerate new ideas in healthcare and cancer diagnosis. 3 Advances in medical imaging and AI have heralded a new era of possibilities in healthcare. 4 Integrating AI into medical data has demonstrated the potential to improve the accuracy and efficiency of timely breast cancer diagnosis, personalize treatment plans, and enhance patients’ quality of life. 5 This potential has led to an increase in studies in the field, necessitating the identification of popular research topics, the status of AI in the field, prominent authors, journals, and sources. Bibliometric indicators are increasingly used to evaluate research performance. 6 One of the primary aims of bibliometric analysis (BA) is to provide quantitative and qualitative insights into the structure, impact, and dynamics of the academic communication environment. 7 In a study conducted by Khairi et al. in 2021, the use and development of deep learning for breast cancer classification were highlighted. 8 Similarly, a bibliometric analysis study by Salod and Singh in 2020, focusing on studies between 2015–2019 on breast cancer prediction using machine learning, interpreted the use of machine learning in breast cancer screening and detection as promising. 9
This study aims to provide a bibliometric overview of studies on AI applications in breast cancer diagnosis. By examining studies written from 2013 to October 2024, the existing literature will be better understood, providing insights into the significance of the field. This will enable academics to see emerging research areas and integrate them into their own studies. Breast cancer is the most frequently diagnosed cancer type in women worldwide and also has a significant place among cancer-related deaths. 10 The increase in the incidence of breast cancer in developed and developing countries increases the importance of research on early diagnosis, new treatment methods, and prognostic factors for this disease. 11 In addition, in recent years, artificial intelligence-supported diagnostic systems have made significant progress in breast cancer diagnosis, and a considerable increase has been observed in academic studies in this field. 12 Therefore, by focusing on breast cancer in our study, we aimed to fill the existing gaps in the scientific literature and contribute to clinical applications in this field.
Methods
This study is a bibliometric analysis of articles on AI applications in breast cancer diagnosis.
Data extraction
A comprehensive search was conducted in the Web of Science database from January 1, 2013, to October 3, 2024. Scopus and PubMed databases were also scanned for additional data, but WOS provided the most comprehensive data, so only this database was used.
Search Query: TS = ((((‘'breast cancer'’) AND (‘'breast cancer detection artifical intelligence'’)) OR (‘'breast cancer detection machine learning'’)) OR (‘'breast cancer detection deep learning'’)) OR (‘’breast cancer detection data mining'’)
Inclusion Criteria: Articles published between 2013 and 2024 in English.
Exclusion Criteria: Publications unrelated to breast cancer, such as AI applications in other cancers or general AI research. Articles include conference papers, review articles, book chapters, retracted publications, editorial materials, data papers, corrections, and articles with incomplete bibliographic data. Data Extraction, Inclusion and Exclusion Criteria are shown in Figure 1.
Data search-flowchart.
The extracted data was loaded into the VOSviewer software, and bibliometric methods were applied to determine the field's focal points and future trends. The Bibliometrix R package was used to obtain the annual number of articles and the thematic map of keywords.
Bibliometric analysis was conducted using the following metrics:
A publication growth analysis was conducted to examine annual publication trends, while a thematic map analysis identified key topics within the field, their strategic importance, and development level over time. The thematic map is visualized in four quadrants: the upper-right quadrant represents the most developed topics, the upper-left shows advanced but less central topics, the lower-left highlights outdated concepts, and the lower-right represents emerging or developing topics.
Citation analysis identified the most influential articles by examining citation counts. Author network analysis evaluated the contributions of authors, institutions, and countries, identifying leading researchers in AI applications for breast cancer diagnosis. Keyword co-occurrence analysis determined the main themes in the field, while collaboration networks assessed partnerships among researchers, institutions, and countries. Journal citation analysis ranked the most productive and impactful journals by publication count and citation impact, and institutional citation analysis identified the most influential institutions in the field.
The bibliometric analysis used in our study was performed with data from the WOS database, which is widely accepted in scientific research. To increase the reliability of the data, the keywords and filters used were clearly stated in the method section, and the analysis was designed to be repeatable. In terms of verification, comparisons were made with trends in previous studies, and the validity of our analysis methods was discussed. In addition, the accuracy of the calculations with VOSviewer and the R Program was ensured.
Results
A total of 1537 articles were retrieved from the Web of Science database following the search between January 1, 2013, and October 3, 2024. The flowchart related to data extraction is shown below.
When the annual growth trend of studies in Figure 2 is examined, it is observed that the number of publications increased by 1.53 times in 2019 compared to 2018, reaching 76 articles. The number of studies on this topic showed a noticeable rise starting in 2019. The number of studies on this topic peaked in 2023 with 365 publications, making it the year with the highest number of publications.
Annual growth of studies.
When the results of the conceptual structure analysis in Figure 3 are examined: In the top-right area, terms like imaging, histopathology, hybrid AI, classification, and support vector machines are seen. These terms represent leading concepts in relevant studies and direct the current research focus. In the top-left area, terms such as metastasis, gene, lymph node, and prediction appear, reflecting less-known, specific, and complex research areas. Genetic mutations play a critical role in breast cancer development, with AI algorithms helping to analyze genetic data to identify risk factors and cancer types. In the bottom-right area, terms like breast cancer, deep learning, machine learning, mammography, and prediction indicate the central role of AI in breast cancer diagnosis. The bottom-left area includes terms like mammography, convolutional, and invasive ductal carcinoma, suggesting the rise of deep learning techniques in this field.
Conceptual structure-thematic map.
The conceptual structure thematic map in Figure 3 helps us understand current scientific trends by visualizing our study's main research areas, topics, and key concepts. This map aims to group the main research themes determined by the bibliometric analysis method on different axes by evaluating the role of artificial intelligence in breast cancer diagnosis and treatment. In particular, it highlights how the field is shaping up and which topics need to be further investigated by revealing the relationships between leading topics, emerging research areas, and specific subtopics in the literature.
When the graph of the country collaboration network analysis results in Table 1 is examined: Evaluating collaborations among countries, the most prominent collaboration is between India and the USA. Additionally, strong collaborations exist between China and the UK. Turkey collaborates with a total of 21 countries. The top five countries with the highest number of collaborations in this field are India (405), China (288), the USA (242), Saudi Arabia (168), and Pakistan (94). As shown in Figure 4, countries such as England (86)
Top 10 most collaborative countries based on co-authorship analysis.
Keyword analysis.
When the graph of the keyword analysis results in Figure 4 is examined: There are a total of 3191 keywords. The 95 most frequently occurring words were selected. In the map, these 95 words are grouped into 7 separate clusters. The most frequently occurring words, in order, are as follows: breast cancer (724), deep learning (504), machine learning (286), mammography (127), classification (120), AI(104), transfer learning (99), CNN (86), feature selection (70), breast cancer diagnosis (70). This indicates that innovative techniques and advanced image processing methods are being applied in this field. The top 10 most frequently occurring words are also presented in Table 2.
Top 10 most common keywords.
When the map of the author citation analysis results in Figure 5 is examined: The network map of 631 authors was divided into 28 clusters, each assigned a different color. The clusters represent groups of authors working on similar topics. According to the author citation analysis results, the top five most cited authors are Nico Karssemeijer (2746), Geert Litjens (2690), Bram Van Ginneken (2504), Mitko Veta (1795), and Hannah Ginmore (970).
Author citation analysis.
Figure 6 presents a comparative view of the top 10 prominent authors in the literature in terms of both productivity (number of publications – blue area) and collaboration intensity (total link strength – red area), in order to examine the relationship between scientific collaborations and productivity levels and to identify influential researcher profiles.
Radar (glyph) chart showing the distribution of authors based on the number of publications and total link strength metrics.
Yu-Dong Zhang, who has the highest number of publications and a strong link strength, stands out as one of the most productive and influential researchers in the network. Although Nico Karssemeijer has a relatively lower number of publications, he draws attention with the highest link strength, indicating that he produces a limited number of publications with high impact and collaboration. Ram Sarkar, T.R. Mahesh, and Robertas Damasevicius display a balanced profile in both metrics, showing consistent performance in terms of both productivity and collaboration. On the other hand, Lei Zhang has lower values in both publication count and link strength compared to the other authors.
Table 3 presents the country citation analysis of the top 10 most cited countries. According to the table, the United States (USA) ranks first with 7511 citations, followed by China (6033), India (5156), the Netherlands (3311), and the United Kingdom (2585). Turkey ranks 16th with 837 citations. India has the highest number of collaborations with 71 countries, followed by China with 61 connections and the USA with 56 connections.
Top 10 most cited countries.
When the journal citation analysis results in Table 4 are examined: This table indicates that the most cited source related to the topic is Scientific Reports, with 2174 citations. It is followed by Medical Image Analysis (1793), IEEE Transactions on Medical Imaging (1582), and IEEE Access (1101). This graph demonstrates that artificial intelligence in diagnosing breast cancer is studied across various disciplines, including computer science, biomedical engineering, and clinical applications. The h-index results of journals are also evaluated in Table 1. Although Biomedical Signal Processing and Control has been analyzed only since 2017, it ranks highest with an h-index of 19, indicating that it has gained significant impact in a relatively short period. Scientific Reports has the highest g-index (43) and total citations (2174), meaning that the most highly cited studies were published in this journal. NP refers to the number of publications analyzed in this study, with IEEE Access having the highest number, with 55 articles included.
H-Index rankings of the Top 10 journals.
Figure 7 the distribution of studies on the use of artificial intelligence applications in breast cancer diagnosis by research areas reveals the highly interdisciplinary nature of the field. Computer Science stands out as the most contributing specific field with a share of 23%, closely followed by Engineering with 21%. In parallel with the importance of imaging technologies, the field of Radiology, Nuclear Medicine, and Medical Imaging accounts for 9%. Oncology represents direct clinical applications with a 5% share, while 4% of the studies fall under other topics in Science and Technology.
Pie chart of research areas.
Figure 8 presents the result of the institution citation analysis; the most cited institution is Radboud University Nijmegen in the Netherlands, with 3048 citations. It is followed by Eindhoven University of Technology (1960), University Medical Center Utrecht (1936), Cairo University (Egypt) (1109), and Case Western Reserve University (987). While Vellore Institute of Technology is the most productive institution with 31 publications, it ranks 50th in citation analysis with 283 citations, indicating high productivity but low impact. With 18 studies on the topic, Radboud University Nijmegen ranks 7th in productivity but stands as the most impactful institution.
Top 10 most cited institutions.
When the results of the bibliographic coupling analysis of the studies in Table 5 are examined: In the field of artifical intelligent applications for breast cancer diagnosis, the top 5 most-cited studies in terms of bibliographic coupling are as follows: Bejnordi's 2017 study ranks first with 1909 citations. This is followed by the studies of Walid Al-Dhabyani (2020) with 808 citations, Thijs Kooi (2017) with 659 citations, Xu (2016) with 601 citations, and Chen (2017) with 509 citations. Table 2 examines the studies with normalized citation counts. The normalized citation count is calculated by dividing the number of citations a publication has received by the total number of citations in its publication year or within its field. The study
Top 10 most cited documents.
In Figure 9, the result of the concept co-occurrence network analysis is presented: The most frequently used words, in order, are breast cancer (1079), model (873), accuracy (823), detection (815), dataset (763), imaging (648), performance (634), teknigue(618), approach (582), and woman (582). These words indicate that research is being conducted to enhance the effectiveness of artificial intelligence, particularly in imaging and data analysis, in breast cancer diagnosis.
Top 10 most frequent terms from consept co-occurence analysis- concept co-occurrence network analysis.
A bibliometric analysis of studies on AI applications in breast cancer diagnosis conducted between January 2013 and October 2024 was performed. Both quantitative and qualitative data were analyzed. The number of studies in the field increased by 1.53 times in 2019, and 2023 was noted as the year with the highest output, with 365 publications. Between 2013 and 2024, there was a 29.4-fold increase in the number of publications.
This growth can be attributed to advancements in AI technologies and their increasing applicability in medicine. According to the findings, it is observed that the number of academic articles in the field of artificial intelligence (AI) increased until 2023 but decreased slightly from 2024 onwards. There are various factors behind this fluctuation. It has allocated significant funds to AI research, increasing academic productivity, especially in deep learning and large language models. 13 The emphasis on the applied aspect of AI instead of fundamental theoretical research in 2024 and beyond may have caused a decrease in new areas of discovery in the academic world. In addition, the high interest in AI in the health field during the pandemic increased the number of academic publications. However, the change in research priorities after the pandemic led to a decrease in academic productivity in specific subfields. 14 Another critical factor is the changes in academic publication processes. AI studies could be published quickly between 2019 and 2023. However, after 2023, more stringent peer review processes may have led to decreased article acceptance rates. 15 Finally, the more significant role of the private sector in AI research compared to academia may have been effective in decreasing academic studies. 16 In the coming years, a new upward trend may be seen with the prominence of interdisciplinary research and topics such as ethical AI.
The global impact of breast cancer and its high mortality rates are driving interest in AI applications. 17 -18 AI's pattern recognition and decision-making capabilities offer promise for detection, diagnosis, personalized treatment, risk assessment, and prevention. 17 Tan et al. highlighted growing interest in AI integration in breast imaging and noted that numerous researchers are examining this integration from various perspectives. 19 The most frequently used terms in concept co-occurrence analysis, such as “AlexNet” and “ResNet,” reflect the application of deep learning in image processing for breast cancer diagnosis. Wang et al. demonstrated in their study on the implementation of deep learning networks in medical image analysis that some studies have developed new networks based on popular existing networks like AlexNet, U-Net, ResNet, VGG, and GoogLeNet. 20
The core themes, including random forest, support vector machines, decision trees, and neural network algorithms, indicate that these are the most commonly used algorithms in this field. 21 When examining international collaborations, the results of the author collaboration network analysis exhibit a distribution consistent with the fundamental principles of Lotka's Law. In other words, it was concluded that a small number of authors wrote a large number of articles, whereas a large number of authors wrote a small number of articles. The country collaboration network analysis result, evaluated in terms of the number of publications, reveal that the distribution of collaborating countries in this field is highly uneven. Citation analyses show that the most cited authors are researchers from the United States and the Netherlands, and it is noteworthy that these authors frequently cite each other's research. When citation analysis is evaluated in terms of countries, it is observed in the upper-layer visualization that countries with new citations or collaborations, such as Turkey, Brunei, Lebanon, Ghana, and Qatar, have shown an increasing trend in this regard since the end of 2022, with its impact intensifying in early 2023. When collaborations and citation analyses are considered, countries with more collaborations tend to have higher citation rates. Regarding author collaboration by country, India appears to be collaborative; however, when scientific impact is considered, the United States is more influential It has been observed that India and China together account for over 45% of the literature in the dataset. This aligns with both countries’ enormous populations and numerous research institutions. Although the effectiveness of studies conducted in the Netherlands is high, its share in the productivity ranking among all countries is 2%, and it ranks fourth in citation analysis. Similarly, Australia holds a 2% share in the productivity ranking but ranks 11th in terms of citation analysis. Despite its lower share in productivity, the Netherlands has been shown to produce the most impactful and widely recognized studies. Turkey has a 1.8% share in productivity.
In keyword analysis, the frequent use of terms such as mammography, ultrasound, and histopathological imaging indicates that artificial intelligence in breast cancer diagnosis focuses on extracting features from breast images and automating the diagnosis process through the application of algorithms. 22 This highlights that the most common application of artificial intelligence technology in breast cancer diagnosis is in the field of histopathological imaging. Over the past eight years, methods related to deep learning, particularly CNN, have demonstrated exceptional performance in breast cancer image classification.8–20 The fields where the most studies have been published are computer science and engineering, which together account for 78% of the publications. Based on the results of the concept co-occurrence network analysis, the following observations can be made: the core components used in diagnosing breast cancer with artificial intelligence are frequently employed together. Deep learning and machine learning models in breast cancer diagnosis have been emphasized. Data mining reflects a process where features selected for the classification of cancer cells are evaluated with appropriate algorithms.
Data sets derived from various imaging techniques, such as ultrasound and histopathology, are frequently analyzed using advanced models like CNN and ResNet-50. The presence CAD systems in this analysis demonstrates their role in breast cancer diagnosis. Studies have shown that when CAD systems are used to analyze ultrasound features, they enhance the diagnostic performance of both inexperienced and experienced physicians. 23 Deep learning models such as AlexNet are trained on digital mammography data, like DDSM, and are used to classify abnormal structures in breast tissue. The use of these concepts underscores the importance of AI-based systems in breast cancer diagnosis and highlights the significance of machine learning in medical image analysis. 24
Limitations
Despite the contributions of this study, it should be noted that only Web of Science was used as the bibliometric data source. Therefore, similar studies could be conducted using other databases. However, even with additional databases, the main results are expected to remain broadly consistent regarding key topics and themes. The articles were collected only up to October 2024. If the collection had extended until December, it is predicted that the growth rate of the field would have increased rather than decreased, although not at the same rate as the number of studies conducted in previous years.
Conclusion
This article presents a detailed bibliometric analysis of the literature on AI applications in breast cancer diagnosis. A 10-year time frame was analyzed using data from the WOS database to gain a detailed perspective on the current and future state of AI algorithms in this field. Trends in existing topics and currently prominent research areas were examined.
The number of publications has increased especially after 2019, reaching its peak in 2023. This growth is attributed to advancements in artificial intelligence technologies and their increasing use in medicine. Deep learning models such as AlexNet and ResNet-50 are particularly common in the classification of breast images. The most frequently occurring keywords and thematic areas focus on histopathological imaging, mammography, ultrasound, and automated diagnosis systems. CNN and other machine learning algorithms are at the core of these studies. In terms of global contributions, India and China produce the highest number of publications, while the United States and the Netherlands stand out in terms of citation impact.
It was observed that countries like the USA, India, and China are leading in terms of productivity and collaboration. Encouraging the use of these technologies in less developed countries could further strengthen international collaborations. Open datasets like UCI could lead to repetitive or similar results in studies; therefore, increasing data diversity by incorporating datasets from diverse patient profiles could enhance the generalizability of models. Along with promoting data diversity, encouraging data sharing could increase participation and collaboration from less productive countries. These findings can help researchers focused on AI applications in breast cancer diagnosis to see the evolving and emerging directions in the field and integrate these innovations into their studies.
Footnotes
Acknowledgments
I would like to express my gratitude to my advisor, Hakan TEKEDERE
Ethical approval
No patients were involved in this research, so the ethical approval is not required. The data sets analyzed during this study are available in the Web of Science core collections.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
