Research trends on AI in breast cancer diagnosis, and treatment over two decades Authors: Alok Singh, Akanksha Singh, Sudip Bhattacharya Published: 2024 ABSTRACT ObjectiveRecently, the integration of Artificial Intelligence (AI) has significantly enhanced the diagnostic accuracy in breast cancer screening. This study aims to deliver an extensive review of the advancements in AI for breast cancer diagnosis and prognosis through a bibliometric analysis.MethodologyTherefore, this study gathered pertinent peer-reviewed research articles from the Scopus database, spanning the years 2000 to 2024. These articles were subsequently subjected to quantitative analysis and visualization through the Bibliometrix R package. Ultimately, potential areas for future research challenges were pinpointed.ResultsThis study analyzes the development of Artificial Intelligence (AI) research for breast cancer diagnosis and prognosis from 2000 to 2024, based on 2678 publications sourced from Scopus. A sharp rise in global publication trends is observed between 2018 and 2023, with 2023 producing 456 papers, indicating intensified academic focus. Leading contributors include ZHENG B, with 36 publications, and institutions like RADBOUD UNIVERSITY MEDICAL CENTER and the IEO EUROPEAN INSTITUTE OF ONCOLOGY IRCCS. The USA leads both in publications (473) and total citations (18,530), followed by India with 289 papers. Co-occurrence analysis shows that “mammography” (3171 occurrences) and “artificial intelligence” (1691 occurrences) are among the most frequent keywords, reflecting core themes. Co-citation network analysis identifies foundational works by authors like Lecun Y. and Simonyan K. in advancing AI applications in breast cancer. Institutional and country-level collaboration analysis reveals the USA’s significant partnerships with China, the UK, and Canada, driving the global research agenda in this field.ConclusionIn conclusion, this bibliometric review underscores the growing influence of AI, particularly deep learning, in breast cancer diagnosis and treatment research from 2000 to 2024. The United States leads the field in publications and collaborations, with India, Spain, and the Netherlands also making significant contributions. Key institutions and journals have driven advancements, with AI applications focusing on improving diagnostic imaging and early detection. However, challenges like data limitations, regulatory hurdles, and unequal global collaboration persist, requiring further interdisciplinary efforts to enhance AI integration in clinical practice. Introduction Generative AI is revolutionizing breast cancer diagnosis by enhancing the accuracy and efficiency of early detection methods. Utilizing advanced algorithms, generative AI models can analyze vast amounts of medical data, including mammograms, ultrasounds, and MRI scans, to identify subtle patterns and anomalies that may indicate the presence of cancer. These models are trained on diverse datasets, enabling them to detect even the most minute changes in breast tissue that might be overlooked by traditional diagnostic techniques. Moreover, generative AI can assist in predicting the progression of the disease and personalizing treatment plans, leading to improved patient outcomes [1]. This technology speeds up the diagnostic process by reducing false positives and negatives. It alleviates the emotional and physical burden on patients, paving the way for more precise and individualized healthcare [2]. Generative AI plays a transformative role in breast cancer diagnosis, as evidenced by various studies published. Research on deep learning and medical image analysis demonstrates how these advanced algorithms can enhance the accuracy of mammogram and histopathological image interpretation, leading to more reliable breast cancer diagnoses [3]. Specific types of deep learning models or machine learning approaches, such as convolutional neural networks (CNNs) or generative adversarial networks (GANs), have played distinct roles in breast cancer diagnosis and prognosis. A study aimed to identify the optimal AI-based method for mitotic figure scoring in breast cancer (BC). Automated mitosis detection was used to assess mitotic counts through three methods: mitotic count per tumor area (MCT), mitotic index (MI), and mitotic activity index (MAI). These methods were then compared with the Nottingham grading system, Ki67 scores, clinicopathologic parameters, and patient outcomes. AI-based scores from all three methods (MCT, MI, and MAI) showed significant correlations with clinicopathologic characteristics and patient survival (P < 0.001) [4]. A thesis regarding breast screening-AI framework integrating multimodal imaging and AI for image analysis revealed enhanced diagnostic efficiency, reduced cognitive workload, and improved clinician satisfaction. It highlighted the transformative impact of intelligent agents in supporting clinical workflows and decision-making. It examined how different communication tones between agents and clinicians—suggestive (non-assertive) versus imposing (assertive)—affect clinician receptivity and performance. Findings suggested that assertive agents can reduce medical errors and increase satisfaction, offering a new approach to adaptive communication in healthcare. This work has advanced human–computer interaction by emphasizing the importance of clinician-centred, intelligent agent design and offers valuable insights for future healthcare innovations. [5] Another study developed a deep learning (DL) model to improve skin cancer classification, addressing class imbalance and enhancing interpretability. Using the HAM10000 dataset, an optimized CNN identified seven cancer types with 82% accuracy and 0.47% loss. An explainable AI (XAI) system with Grad-CAM methods provided insights into model decisions, aiding early diagnosis and reducing clinical workload. An end-to-end smart healthcare system was also proposed via an Android app, supporting dermatologists in timely, informed decision-making. [6] Deep learning has significantly advanced breast cancer detection with computer-aided diagnosis, but most work focuses on 2D images. Addressing the challenges of 3D MRI, such as limited data and annotations, this study introduces a two-stage deep multiple instances learning framework. Using only global labels, it enables whole-volume and slice-level classification and 3D lesion localization via heatmaps. Results show competitive performance with the state-of-the-art, and heatmap assessments demonstrate effective malignant lesion identification [7]. A study employed artificial intelligence (AI) to evaluate the prognostic value of tumour-infiltrating lymphocytes (TILs) in a large, well-defined cohort of luminal breast cancer (BC) patients. Findings indicated that AI-based analysis of TIL counts and spatial distribution offers prognostic insight for early-stage luminal BC. The use of AI algorithms enables a detailed assessment of TILs in whole-slide images, surpassing the limitations of manual examination [8]. Another study proposed a multimodal framework that processed all available MRI data to reach a diagnosis, rather than relying on a single volume, emulating the radiologists' workflow. The framework comprised a 3D convolutional neural network for each modality, with predictions combined using a late fusion strategy based on Dempster-Shafer theory. The results highlighted the most relevant modalities needed for accurate diagnosis, aligning with clinical practice, and showed that combining multiple modalities led to better overall results than using them individually. [9] Studies on Generative Adversarial Networks (GANs) reveal their utility in data augmentation for mammography, allowing for creating more comprehensive training datasets that improve AI model performance. [10, 11] Additionally, AI-based image generation techniques are being used to produce synthetic breast cancer images, which help refine diagnostic tools and increase their robustness. [12, 13] These advancements underscore the significant impact of generative AI in improving early detection, diagnostic accuracy, and personalized treatment plans for breast cancer patients. [14] With this background, this bibliometric analysis aims to find out the research trends on generative AI in breast cancer diagnosis, prognosis and treatment over two decades. Materials and methods Methodology and data sources Preliminary-planning In the preliminary planning stage, search queries were pinpointed and organized as illustrated in Table 1. These queries were divided into two categories.Primary search terms.Combinations of primary search terms with breast cancer-related items focused on diagnosis and treatment of breast cancer.Table 1List of keywords( TITLE-ABS-KEY ( ai ) OR TITLE-ABS-KEY ( “artificial intelligence” ) OR TITLE-ABS-KEY ( “regenerative ai” ) OR TITLE-ABS-KEY ( “responsible ai use” ) OR TITLE-ABS-KEY ( “machine learning” ) OR TITLE-ABS-KEY ( “convoluted neural network” ) OR TITLE-ABS-KEY ( “ethical dilemna in AI” )ANDTITLE-ABS-KEY ( “benign breast lesions treatment” ) OR TITLE-ABS-KEY ( “benign breast lesions diagnosis” ) OR TITLE-ABS-KEY ( mammography ) OR TITLE-ABS-KEY ( “breast malignancy” ) OR TITLE-ABS-KEY ( “breast lump” ) OR TITLE-ABS-KEY ( “brest cancer treatment” ) OR TITLE-ABS-KEY ( "brest cancer diagnosis" ) Primary search terms. Primary search terms. Combinations of primary search terms with breast cancer-related items focused on diagnosis and treatment of breast cancer. Combinations of primary search terms with breast cancer-related items focused on diagnosis and treatment of breast cancer. Table 1List of keywords( TITLE-ABS-KEY ( ai ) OR TITLE-ABS-KEY ( “artificial intelligence” ) OR TITLE-ABS-KEY ( “regenerative ai” ) OR TITLE-ABS-KEY ( “responsible ai use” ) OR TITLE-ABS-KEY ( “machine learning” ) OR TITLE-ABS-KEY ( “convoluted neural network” ) OR TITLE-ABS-KEY ( “ethical dilemna in AI” )ANDTITLE-ABS-KEY ( “benign breast lesions treatment” ) OR TITLE-ABS-KEY ( “benign breast lesions diagnosis” ) OR TITLE-ABS-KEY ( mammography ) OR TITLE-ABS-KEY ( “breast malignancy” ) OR TITLE-ABS-KEY ( “breast lump” ) OR TITLE-ABS-KEY ( “brest cancer treatment” ) OR TITLE-ABS-KEY ( "brest cancer diagnosis" ) List of keywords The primary search terms encompassed “AI,” “artificial intelligence,” “regenerative AI,” “responsible AI use,” “Machine Learning,” and a variety of supervised and unsupervised algorithms, as shown in Table 1. The secondary search terms, also presented in Table 1, included combinations related to “benign breast lesions treatment,” “benign breast lesions diagnosis,” “mammography,” “breast malignancy,” “breast lump,” “breast cancer treatment,” and “breast cancer diagnosis.” A subset of crucial search queries and their combinations were chosen based on their relevance to AI's application in breast cancer diagnosis and treatment research. We broadened our literature review to encompass combinations of key search terms related to breast cancer, focusing on diagnosis and treatment. By including criteria for breast cancer and microarray gene expression with terms like artificial intelligence (AI) and machine learning (ML), our goal was to investigate how AI and ML are applied in breast cancer research utilizing microarray gene expression data. This data is crucial for understanding the roles of gene biomarkers in the onset and progression of breast cancer. AI and ML techniques help identify the most significant gene biomarkers, aiding in the development of classification and deep learning models for predicting and classifying various stages of the disease [15]. The integration of gene microarray data with AI has shed light on the evolving role of AI in breast cancer diagnosis and treatment over the past two decades [16]. Moreover, specific research questions, outlined in Table 2, were formulated to provide a comprehensive overview of the knowledge structure and the bibliometric and statistical methods used to evaluate AI research in breast cancer detection and survival prediction from 2000 to 2024.Table 2Mapping of research questions with knowledge structure, bibliometric and statistical techniquesS.noResearch QuestionsKnowledge structure coveredBibliometric techniques1What are the publishing and citation trends of the research publication in AI for Breast Cancer Diagnosis and Treatment?Intellectual structureAnnual Scientific Production, Average citation per year2Who are the most contributing authors, journals, organizations, and countries in AI Breast Cancer Diagnosis and Treatment?Intellectual structureThree field plot, Most relevant Authors, Co-citation, and Coupling3What are the publication patterns and most frequently used keywords of the articles published in AI for Breast Cancer Diagnosis and Treatment?Conceptual structureNetwork Analysis, Factorial analysis, Thematic mapping, and Thematic evolution4What are the collaboration networks of AI research in Breast Cancer Diagnosis and Treatment?Social structureAuthors collaboration network, Institution collaboration network, and Country Collaboration WorldMAP5What are the thematic trends of the Application of AI in Breast Cancer Diagnosis and Treatment research and development?Conceptual structureThematic mapping, Thematic evolution, and Factorial analysis6What are the main open areas of challenges and the corresponding solutions for future research work in AI for breast cancer research?Conceptual structureThematic mapping, Thematic evolution, and Factorial analysis Table 2Mapping of research questions with knowledge structure, bibliometric and statistical techniquesS.noResearch QuestionsKnowledge structure coveredBibliometric techniques1What are the publishing and citation trends of the research publication in AI for Breast Cancer Diagnosis and Treatment?Intellectual structureAnnual Scientific Production, Average citation per year2Who are the most contributing authors, journals, organizations, and countries in AI Breast Cancer Diagnosis and Treatment?Intellectual structureThree field plot, Most relevant Authors, Co-citation, and Coupling3What are the publication patterns and most frequently used keywords of the articles published in AI for Breast Cancer Diagnosis and Treatment?Conceptual structureNetwork Analysis, Factorial analysis, Thematic mapping, and Thematic evolution4What are the collaboration networks of AI research in Breast Cancer Diagnosis and Treatment?Social structureAuthors collaboration network, Institution collaboration network, and Country Collaboration WorldMAP5What are the thematic trends of the Application of AI in Breast Cancer Diagnosis and Treatment research and development?Conceptual structureThematic mapping, Thematic evolution, and Factorial analysis6What are the main open areas of challenges and the corresponding solutions for future research work in AI for breast cancer research?Conceptual structureThematic mapping, Thematic evolution, and Factorial analysis Mapping of research questions with knowledge structure, bibliometric and statistical techniques Data collection During the data collection phase, we systematically searched academic articles in the Scopus core collection from January 1, 2000, to June 20, 2024, focusing on the application of artificial intelligence in breast cancer diagnosis and prognosis. The keywords employed for data retrieval are listed in Table 1. Additionally, English-language research articles and review papers were included in the study. This search yielded 2,690 academic publications from Scopus for analysis. Data refinement During the data refinement stage, we filtered the publications obtained from Scopus using specific exclusion criteria. We excluded books, editorials, letters, conference papers, and non-english academic works from our systematic bibliometric review. After this initial filtering, we removed duplicates from the remaining list of 2690 publications, resulting in a final total of 2678 articles. Data extraction We extracted metadata from Scopus in the form of a CSV bibliographic information file. The exported data included: (a) authors/editors, (b) full names of authors, (c) titles, (d) sources, (e) authors’ keywords, (f) keywords plus, (g) abstracts, (h) authors’ affiliations, (i) corresponding authors' affiliations, (j) cited references, (k) total citations, (l) highly cited papers, (m) usage counts, (n) publication years, (o) DOIs, (p) subject categories, (q) author identifiers, (r) languages, and (s) funding agencies. Bibliometric analysis Bibliometric analysis serves as an objective method for researchers to catalogue, access, and evaluate extensive collections of publications, offering a detailed overview of recent trends in scientific literature within a specific field or research area. In this research, we conduct a bibliometric analysis of publications concerning the development of AI research in breast cancer diagnosis and prognosis from 2000 to 2024, addressing the six primary questions presented in Table 2. We employed biblioshiny for bibliometric data analysis to illustrate publication patterns and research trends in applying artificial intelligence to breast cancer diagnosis and prognosis [17]. Furthermore, we aim to statistically investigate and evaluate the scientific knowledge structure through this bibliometric analysis. The fundamental knowledge framework of a research field comprises three components:Conceptual structure: Central themes and trends in the literature of a specific research area.Intellectual structure: The impact of an author’s work within the scientific community.Social structure: Interactions among authors, institutions, and countries. Conceptual structure: Central themes and trends in the literature of a specific research area. Conceptual structure: Central themes and trends in the literature of a specific research area. Intellectual structure: The impact of an author’s work within the scientific community. Intellectual structure: The impact of an author’s work within the scientific community. Social structure: Interactions among authors, institutions, and countries. Social structure: Interactions among authors, institutions, and countries. Initially, the conceptual structure is examined statistically using thematic mapping [18], and keyword co-occurrence networks [19]. Subsequently, the intellectual knowledge structure is evaluated through co-citation network analysis. Lastly, the social knowledge structure is scrutinized based on the collaboration network and collaboration world map. By analysing these conceptual, intellectual, and social structures, we aim to comprehend the knowledge framework of AI applications in breast cancer diagnosis and prognosis over the past two decades. This analysis will highlight current achievements and identify future challenges in implementing AI for breast cancer diagnosis and prognosis. Results Analysis of annual scientific production The number of publications from 2000 to 2024 illustrates the development of research and artificial intelligence (AI) trends for breast cancer diagnosis and prognosis. This study utilizes the Scopus database to extract 2678 academic publications from this period, as detailed in Fig. 1, which depicts the annual variations in scientific contributions within this research domain over the specified timeframe. The analysis of annual scientific production indicates that global publication trends in artificial intelligence for breast cancer diagnosis prognosis, and treatment reached their zenith between 2018 and 2023, with 2023 emerging as the most prolific year, accounting for 456 scientific publications. Consequently, the marked increase in international academic literature over the past six years (2018 to 2023) reflects an escalating intensity of research in artificial intelligence for breast cancer diagnosis, prognosis and treatment. This suggests that Artificial Intelligence research in this field has garnered significant attention from the academic community during this period.Fig. 1Annual scientific production Annual scientific production Analysis of most relevant authors The following section showcases prominent researchers in Artificial intelligence for breast cancer diagnosis, prognosis and treatment, evaluated by their publication records and the significance of their contributions. Table 3 presents the top ten prolific authors, detailing their publication count, total citations, and h-index. ZHENG B leads with 36 publications, followed by LI J with 29, KARSSEMEIJER N with 27, and both MANN RM and WANG Y with 25. CHAN H-P and NISHIKAWA RM have 19 publications each, SECHOPOULOS I has 18, and MOY L and HOUSSAMI N each have 16.Table 3Most relevant and prolific authorsElementNPh_indexTCKarssemeijer N27162133Zheng B3615818LI J2913933Mann RM25131449Chan H-P1911986Moy L1611780Sechopoulos I18111206Wang Y2511505Houssami N1610612Nishikawa RM19101370 Table 3Most relevant and prolific authorsElementNPh_indexTCKarssemeijer N27162133Zheng B3615818LI J2913933Mann RM25131449Chan H-P1911986Moy L1611780Sechopoulos I18111206Wang Y2511505Houssami N1610612Nishikawa RM19101370 Most relevant and prolific authors In terms of citation impact, KARSSEMEIJER N ranks highest with 2133 citations, while MANN RM and SECHOPOULOS I have 1499 and 1206 citations, respectively. The h-index indicates that KARSSEMEIJER N and ZHENG B are the most influential authors, with h-indices of 16 and 15, respectively. They are followed by LI J and MANN RM, each with an h-index of 13, and CHAN H-P, MOY L, SECHOPOULOS I, and WANG Y, each with an h-index of 11. Consequently, the table indicates that ZHENG B, who has the most publications, is a prominent figure in artificial intelligence research for diagnosing and treating breast cancer. Analysis of most relevant organizations Table 4 highlights the leading organizations in artificial intelligence research focused on breast cancer diagnosis, prognosis and treatment. Table 4 identifies the five most prolific institutions, with RADBOUD UNIVERSITY MEDICAL CENTER leading with 91 publications. It is followed closely by IEO EUROPEAN INSTITUTE OF ONCOLOGY IRCCS with 90 publications, UNIVERSITY OF CALIFORNIA with 88 publications, LUND UNIVERSITY with 85 publications, and YONSEI UNIVERSITY COLLEGE OF MEDICINE with 69 publications. Figure 2 uses a three-field plot diagram to illustrate the pattern of authors’ publications in different related topics and journals.Table 4Most relevant organizationsInstitution ArticlesRadboud university medical center91IEO European institute of oncology IRCCS90University of California88Lund university85Yonsei university college of medicine69University of Michigan68University of Pittsburgh58Massachusetts general hospital56The university of Sydney56University of Oklahoma54Fig. 2Three field plot Table 4Most relevant organizationsInstitution ArticlesRadboud university medical center91IEO European institute of oncology IRCCS90University of California88Lund university85Yonsei university college of medicine69University of Michigan68University of Pittsburgh58Massachusetts general hospital56The university of Sydney56University of Oklahoma54 Most relevant organizations Three field plot Analysis of Country scientific production Table 5 lists the ten leading countries involved in artificial intelligence research for breast cancer diagnosis, prognosis and treatment. Table 5 provides data on the total number of published articles in this field, including total citations (TC) and the average citations per article. According to Table 5, only the USA and India have published over 250 papers on AI for breast cancer diagnosis and treatment from 2000 to 2024. The USA is the leading country in terms of scientific productivity, with 473 publications, followed by India, with 289 publications.Table 5Country scientific productionCountryArticlesTotal citationsAverage article citationsUSA47318,53039.20China191384020.10Netherlands48309264.40India289308610.70Italy89218124.50United Kingdom88201322.90Korea63197231.30France242124242.80Spain214120633.50Canada46110324.00 Table 5Country scientific productionCountryArticlesTotal citationsAverage article citationsUSA47318,53039.20China191384020.10Netherlands48309264.40India289308610.70Italy89218124.50United Kingdom88201322.90Korea63197231.30France242124242.80Spain214120633.50Canada46110324.00 Country scientific production Most preferred periodicals Bradford's Law suggests that a small number of journals are central to a specific research field. [20] As illustrated in Fig. 3, the ten journals highlighted in Table 6 form this core group, accounting for approximately one-third of the entire collection's documents. Table 6 provides details on these top ten periodicals, each contributing one or more articles to our bibliographic collection. Notably, RADIOLOGY is the most preferred journal, with 57 articles, followed by EUROPEAN RADIOLOGY and IEEE TRANSACTIONS ON MEDICAL IMAGING, with 40 and 32 articles, respectively. Regarding the H-index, which measures the number of published articles (h) that have been cited at least h times, IEEE TRANSACTIONS ON MEDICAL IMAGING leads with an H-index of 24 and an impressive 4,254 total citations. RADIOLOGY follows with an H-index of 20 and 1,781 total citations. Other prominent journals in the field include ACADEMIC RADIOLOGY (H-index = 17, TC = 694), EUROPEAN RADIOLOGY (H-index = 17, TC = 809), COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE (H-index = 16, TC = 2,162), MEDICAL PHYSICS (H-index = 16, TC = 1,255), COMPUTERS IN BIOLOGY AND MEDICINE (H-index = 15, TC = 1,058), EUROPEAN JOURNAL OF RADIOLOGY (H-index = 15, TC = 723), JOURNAL OF DIGITAL IMAGING (H-index = 14, TC = 865), and MEDICAL IMAGE ANALYSIS (H-index = 13, TC = 2,873). These journals are among the most significant in publishing research on AI in breast cancer detection and survival predictions.Fig. 3(Bradford law)Table 6Top ten journals in this domainElementh_indexTCNPIEEE transactions on medical imaging24425432Radiology20178157Academic radiology1769431European radiology1780940Computer methods and programs in biomedicine16216226Medical physics16125531Computers in biology and medicine15105827European journal of radiology1572331Journal of digital imaging1486529Medical image analysis13287316 (Bradford law) Table 6Top ten journals in this domainElementh_indexTCNPIEEE transactions on medical imaging24425432Radiology20178157Academic radiology1769431European radiology1780940Computer methods and programs in biomedicine16216226Medical physics16125531Computers in biology and medicine15105827European journal of radiology1572331Journal of digital imaging1486529Medical image analysis13287316 Top ten journals in this domain Analysis of the highly cited research publications in AI for breast cancer diagnosis and treatment The ten most frequently cited research publications in the field of AI and breast cancer, specifically within the analyzed collection and published between 2000 and 2024, are listed in Table 7. For instance, Ahmad Hosny et al. authored an article titled 'Artificial intelligence in radiology’ in ‘Nature Reviews Cancer’, which is the highest locally cited publication with 1887 total citations and a normalized TC of 30.60 [21]. The second most influential paper, with 1386 total citations and a normalized TC of 46.06, is “International evaluation of an AI system for breast cancer screening” by Mickinney SM et al., published in ‘Nature’. Desantis Ce’s article 'Breast cancer statistics, 2017, racial disparity in mortality by State’ also garnered 1218 total citations and a normalized TC of 18.76 [22]. Lastly, BI WL et al.’s publication 'Artificial intelligence in cancer imaging: Clinical challenges and applications' achieved 1013 total citations and a normalized TC of 28.7 [23]. Consequently, as depicted in Table 7, these authors are leading contributors to AI research in breast cancer diagnosis and treatment from 2000 to 2024.Table 7Highly cited research publications in AI for breast cancer diagnosis and treatmentPaperDOITotal CitationTC per YearNormalized TCHosny A, 2018, Nat Rev Cancer10.1038/s41568-018-0016-51887269.5730.60Mckinney SM, 2020, Nature10.1038/s41586-019–1799-61386277.2046.06Desantis CE, 2017, Ca Cancer J Clin10.3322/caac.214121218152.2518.76BI Wl, 2019, Ca Cancer J Clin10.3322/caac.215521013168.8328.07Castelvecchi D, 2016, Nature10.1038/538020a928103.1114.85Lopes R, 2009, Med Image Anal10.1016/j.media.2009.05.00381350.8115.48Ricci E, 2007, IEEE Trans Med Imaging10.1109/TMI.2007.89855175041.6714.31Kooi T, 2017, Med Image Anal10.1016/j.media.2016.07.00774292.7511.43Shen L, 2019, Sci Rep10.1038/s41598-019–48995-4607101.1716.82Tang J, 2009, IEEE Trans Inf Technol Biomed10.1109/TITB.2008.200944154934.3110.45 Table 7Highly cited research publications in AI for breast cancer diagnosis and treatmentPaperDOITotal CitationTC per YearNormalized TCHosny A, 2018, Nat Rev Cancer10.1038/s41568-018-0016-51887269.5730.60Mckinney SM, 2020, Nature10.1038/s41586-019–1799-61386277.2046.06Desantis CE, 2017, Ca Cancer J Clin10.3322/caac.214121218152.2518.76BI Wl, 2019, Ca Cancer J Clin10.3322/caac.215521013168.8328.07Castelvecchi D, 2016, Nature10.1038/538020a928103.1114.85Lopes R, 2009, Med Image Anal10.1016/j.media.2009.05.00381350.8115.48Ricci E, 2007, IEEE Trans Med Imaging10.1109/TMI.2007.89855175041.6714.31Kooi T, 2017, Med Image Anal10.1016/j.media.2016.07.00774292.7511.43Shen L, 2019, Sci Rep10.1038/s41598-019–48995-4607101.1716.82Tang J, 2009, IEEE Trans Inf Technol Biomed10.1109/TITB.2008.200944154934.3110.45 Highly cited research publications in AI for breast cancer diagnosis and treatment Conceptual knowledge structure analysis Analysis of the keywords In this part of our study, we employ keyword and co-occurrence analyses to explore the latest research trends and advancements in using artificial intelligence for the diagnosis, prognosis and treatment of breast cancer. Our objective is to pinpoint research gaps and predict future directions in this domain.” The primary keywords are illustrated in Fig. 4, with “mammography” (3171 occurrences) being the most frequent, followed by “female” (1739), “artificial intelligence” (1691), “human” (1483), and “breast cancer” (1444).Fig. 4Most relevant words Most relevant words Additionally, the relationship between AI and breast cancer diagnosis, prognosis and treatment research is depicted using the word growth graph in Fig. 5 and Table 8. This graph shows a progressive increase in the annual occurrence of key terms related to AI tools for early breast cancer diagnosis, such as “breast cancer,” “deep learning (DL),"” “machine learning (ML),” “feature selection,” and “classification.” Notably, terms like “breast cancer,” “breast neoplasm,” “machine learning,” and “deep learning” have exhibited more dynamic growth compared to others.Fig. 5Words frequency over timeTable 8Words frequency over timeYearMammographyFemaleArtificial intelligenceHumanBreast cancerMachine learningHumansArticleDeep learningBreast neoplasms2000632120210220012110126406505200232132287087072003411633129010110920046023431817016160132005993768312102930024200615871107492715450045200721595142653817270059200825011516878432888306920093041432079760411110509020243171173916911483144410621012973892790 Words frequency over time Table 8Words frequency over timeYearMammographyFemaleArtificial intelligenceHumanBreast cancerMachine learningHumansArticleDeep learningBreast neoplasms2000632120210220012110126406505200232132287087072003411633129010110920046023431817016160132005993768312102930024200615871107492715450045200721595142653817270059200825011516878432888306920093041432079760411110509020243171173916911483144410621012973892790 Words frequency over time For example, the cumulative occurrence of the top 10 keywords includes “mammography” (6), “female” (3), “artificial intelligence” (2), “human” (1), “breast cancer” (2), “machine learning” (0), “humans” (2), “article” (1), “deep learning” (0), and “breast neoplasms” (2). In contrast, by 2024, the keywords with the highest increase in occurrences from 2000 to 2024 were “mammography” (3171), “female” (1739), “artificial intelligence” (1691), “human” (1483), “breast cancer” (1444), “machine learning” (1062), “humans” (1012), “article” (973), “deep learning” (892), and “breast neoplasms” (790). Furthermore, we analysed the co-occurrence of authors’ keywords using biblioshiny. The co-occurrence network helps understand the thematic areas covered within the research field and identify the most critical and current issues. It also provides insights into the evolution of these issues over time. The results of the co-occurrence network analysis are presented in Fig. 6. In this figure, the node size (represented by a dot) indicates the number of occurrences of each keyword.Fig. 6Co-occurrence network Co-occurrence network For instance, “mammography” appears as the largest node, confirming its status as the most frequent keyword. Figure 7 reveals that other frequently occurring author keywords include “human, ” “artificial intelligence, ” “female, ” “breast neoplasms, ” “article, ” “breast tumour, ” “machine learning, ” “diagnostic imaging, ” “deep learning, ” and “sensitivity, ” all ranking highly after “mammography. ”Fig. 7Authors keyword Authors keyword Keywords evolution trends Using a clustering algorithm on the keyword network allows for the identification of various themes within a given field. Each cluster or theme can be depicted on a specific type of graph known as a strategic or thematic map [18]. In this map, each bubble signifies a network cluster. The name of the bubble corresponds to the word with the highest frequency within the cluster. The size of the bubble is proportional to the frequency of the cluster’s word occurrences, and its position is determined based on the cluster's Callon centrality and density. Callon centrality indicates the significance of the theme within the entire research area, while Callon density measures the theme's development. Consequently, thematic maps were generated to display the evolution of keyword trends, as illustrated in Fig. 8C.Fig. 8A: Motor Themes. B: Emerging/ Declining Theme. C: Thematic map A: Motor Themes. B: Emerging/ Declining Theme. C: Thematic map The thematic map is divided into four quadrants: The upper right quadrant highlights thematic keywords associated with motor themes, indicating well-developed topics related to the application of AI in breast cancer diagnosis, prognosis and treatment research. The second quadrant features niche themes that exhibit strong internal development. The third quadrant contains thematic keywords representing weakly developed, emerging, or declining themes. Lastly, the fourth quadrant includes thematic keywords related to basic and transversal themes with minimal internal development. Figure 8A illustrates a thematic analysis of data from 2000–2024, revealing that Human, AI, Breast neoplasm, Brest cancer, and diagnostic imaging are both highly developed and crucial to the conceptual framework of the research area (AI for Breast Cancer Diagnosis and Treatment).Conversely, the themes of diseases, medical imaging, computer-aided diagnostics, learning systems, and image segmentation are either emerging or fading, with limited internal advancement. These areas hold marginal significance in the advancement of the specific scientific discipline. Figure 8B. The thematic development of keywords from 2000 to 2024 is examined using the keyword thematic map and Sankey diagram presented in Fig. 9. Analysis of these visualizations reveals that between 2000 and 2021, research primarily concentrated on using computer-aided diagnostics for identifying and categorizing malignancies, particularly breast cancer. In contrast, the past two years have seen a shift towards leveraging artificial intelligence and support vector machines, significantly enhancing early breast cancer detection and prognosis.Fig. 9Sankey diagram Sankey diagram Analysis of intellectual knowledge structure Analysis of Co-citation Analysis of co-citation is a pivotal method in bibliometrics for identifying relationships between nodes that represent authors or documents, thus illustrating the intellectual structure of a specific research field [19]. This technique involves recognizing the co-citation of two papers or authors when both are cited by a third entity. In this context, nodes symbolize the co-cited documents, while the edges between them denote instances of co-citation. The size of a node correlates with the frequency of document occurrence, with more frequently occurring papers having larger nodes. Similarly, the thickness of the edges corresponds to the frequency of co-occurrence, with more frequently co-occurring records having thicker edges. As depicted in Fig. 10, several research papers—such as those by Lecun (2015–1), Simonyan (2014), Xie (2016), Beura (2015), Hanley, Ragab (2019), Jiao (2016), Dhungel. (2017), Goodfellow. (2016), Moreira (2012–2), Ribli. (2018), Krizhevsky (2012), Ronneberger (2015), Vijayarajeswari (2019), Geras (2019–2), Kingma (2014), He (2016), Yala (2019–2), and Liberman—have been cited by other documents and co-cited by many source documents in the dataset. Additionally, authors such as McKinney (2020–1), Freeman (2021), Salim (2020), Houssami, Gillies (2016), Yala (2019–1), Rodriguez-Ruiz (2019–2, 2019–1), Geras (2019–1), McKinney (2020–2), Fenton (2007), and Kooi (2017), along with Lecun (2015–2), have also been co-cited by other source documents. The research fields associated with the records are represented by the colours of the nodes in the co-citation network.Fig. 10Co-citation network Co-citation network Analysis of social knowledge structure. Authors’ collaboration network analysis The analysis of the authors' collaboration network uncovers interaction patterns among them [24]By setting a threshold of at least 10 edges, we depicted the global collaboration of authors from around the world. Figure 11 highlights the partnerships among the fourteen most prolific authors in the dataset. These fourteen authors collaborated strongly with others, each having co-authored at least ten publications. The thickness of the edges indicates the strength of the associations between authors, while the size of the nodes reflects the number of articles they co-authored.Fig. 11Collaboration network of authors Collaboration network of authors For instance, in the brown-coloured research path, Zheng B, Danala G, Heidari M, and Qiu Y published more papers than other authors. Similarly, Chan H-P and Helvie Ma, represented in the yellow-coloured research path, had a higher number of publications in this research field. Lastly, Yang Y and Nishikawa Rm in the green-coloured research field published more articles together than other authors in the dataset within this field. Institution collaboration network analysis The analysis of the institutional collaboration network illustrates the interactions between institutions [24]. We used a threshold of two or more connections to depict the global collaborations among institutions worldwide. The thickness of the connections indicates the strength of associations, while the size of the nodes represents the number of collaborative articles. Figure 12 highlights the collaboration among the top collaborating institutions within the dataset. Notably, the University of Pennsylvania (64.8), Karolinska University Hospital (56.6), Karolinska Institute (46.6), and University of California (42.2) exhibit the highest betweenness centrality among all universities.Fig. 12Collaboration network of institutions Collaboration network of institutions Collaboration world map analysis Figure 13 depicts the collaborative network among countries, emphasizing interactions between different nations. We established a minimum threshold of ten edges to represent global partnerships. According to Table 9, the USA had notable collaborations with China (47), Canada (31), the UK (30), Australia (22), and the Netherlands (20) in AI research focused on breast cancer diagnosis, prognosis and treatment. Furthermore, Saudi Arabia exhibited a strong partnership with Pakistan, with 19 collaborations in this research domain.Fig. 13Pictorial representation of the countrywide collaboration using the country collaboration mapTable 9Highly collaborating countries in AI for breast cancer diagnosis and treatment represented by collaboration World MAPFromToFrequencyUSAChina47USACanada31USAUnited kingdom30USAAustralia22USANetherlands20Saudi ArabiaPakistan19ItalyUnited kingdom16United kingdomSpain16IndiaSaudi Arabia14NetherlandsGermany14 Pictorial representation of the countrywide collaboration using the country collaboration map Table 9Highly collaborating countries in AI for breast cancer diagnosis and treatment represented by collaboration World MAPFromToFrequencyUSAChina47USACanada31USAUnited kingdom30USAAustralia22USANetherlands20Saudi ArabiaPakistan19ItalyUnited kingdom16United kingdomSpain16IndiaSaudi Arabia14NetherlandsGermany14 Highly collaborating countries in AI for breast cancer diagnosis and treatment represented by collaboration World MAP Discussion Artificial intelligence (AI) is continually transforming the way humans operate and has been integrated into various sectors such as agriculture, manufacturing, and advanced healthcare [25–27]. For instance, since the introduction of AI for diagnosing and treating breast cancer, numerous academics, scientists, and researchers have conducted significant experiments utilizing various deep learning (DL) technologies for these purposes. [28] Despite this progress, a comprehensive evaluation of DL applications in breast cancer diagnosis and treatment from a bibliometric standpoint has been lacking. Specifically, existing literature has not adequately addressed six key areas such as:1) Trends in publishing and citations for AI research in breast cancer2) Leading contributors in terms of authors, journals, organizations, and countries3) Frequently used keywords and publication patterns in AI-related breast cancer research4) The collaboration networks within this research domain5) The thematic trends in AI applications for breast cancer diagnosis and treatment6) Challenges and potential solutions for future research in this field 1) Trends in publishing and citations for AI research in breast cancer 1) Trends in publishing and citations for AI research in breast cancer 2) Leading contributors in terms of authors, journals, organizations, and countries 2) Leading contributors in terms of authors, journals, organizations, and countries 3) Frequently used keywords and publication patterns in AI-related breast cancer research 3) Frequently used keywords and publication patterns in AI-related breast cancer research 4) The collaboration networks within this research domain 4) The collaboration networks within this research domain 5) The thematic trends in AI applications for breast cancer diagnosis and treatment 5) The thematic trends in AI applications for breast cancer diagnosis and treatment 6) Challenges and potential solutions for future research in this field 6) Challenges and potential solutions for future research in this field This study aims to address the existing gap by examining the knowledge framework of artificial intelligence in breast cancer diagnosis and treatment through systematic bibliometric review methods. It emphasizes key research trends, publication patterns across various countries and journals, author contributions and networks, and global collaborations in the application of AI for diagnosing and treating breast cancer. The United States has led the publication of research articles on AI for breast cancer diagnosis and treatment, with India and Spain also making significant contributions. Although the USA has the highest global influence based on total citation metrics, the Netherlands stands out in terms of average citations per article in this field. This may be due to Dutch research institutions are renowned for their focus on producing high-quality, impactful studies on AI rather than simply increasing publication volume. As an example, recently they have lay out the theoretical framework, purpose, and structure of a novel AI consortium cardiovascular artificial intelligence (AI) research namely CVON-AI consortium. [29] Additionally, the USA exhibits strong collaborative ties with China, the United Kingdom, Australia, and the Netherlands. Prominent institutions such as Radboud University Medical Center and the IEO European Institute of Oncology IRCCS have been pivotal in AI research for breast cancer diagnosis and treatment from 2000 to 2024. The IEEE Transactions on Medical Imaging is the most influential and preferred journal for researchers in this area, garnering the highest citations during this period. The dominance of institutions like Radboud University Medical Center, IEO European Institute of Oncology, and the University of California in AI research for breast cancer diagnosis and treatment reflects their advanced research infrastructure, interdisciplinary collaboration, and consistent government and private funding. These institutes are renowned for pioneering cancer research and adopting cutting-edge technologies, including AI, to enhance diagnostic accuracy and treatment outcomes. Countries like the USA, China, and India are leading contributors due to their vast populations, high cancer incidence, and significant investments in healthcare technology. The USA's leadership stems from robust funding from agencies like the National Institutes of Health (NIH). China and India are rapidly expanding their AI research capabilities to meet increasing healthcare demands. The correlation between these institutions and countries is driven by the need for scalable, precise cancer care solutions that align with their public health priorities and technological advancements [30]. Regarding most preferred periodicals it was observed that a small number of journals are central to a specific research field. Bradford's Law, which explains the unequal distribution of journal articles across research fields, is widely referenced in bibliometrics, yet its origins are subject to debate. Samuel C. Bradford first introduced this principle in 1934, but some scholars argue that its conceptual roots may predate Bradford's work, connecting it to early ideas in the distribution of information. Despite this uncertainty, the law's practical application in determining core journals for specific subjects remains unquestioned [31]. Moreover, Bradford's Law can be seen as a specific instance of the Pareto distribution. Just as the Pareto principle suggests that 80% of outcomes are derived from 20% of causes, Bradford’s Law similarly highlights that a small proportion of journals account for a large share of important research articles. This correlation underscores the inherent inequality in information dissemination across academic journals. [32] Based on our bibliometric analysis, ZHENG B stands out as the most prolific author, frequently publishing in the field of AI for breast cancer research. Conversely, KARSSEMEIJER N emerges as one of the most impactful authors, boasting 2133 citations and an H-index of 16. The work by Hosny A, Parmar C, et al. (2018) addresses the challenges in clinical implementation and provides insights on advancing the domain, making it the most globally cited and influential article in AI for breast cancer diagnosis and prognosis research from 2000 to 2024. A more nuanced critique of the top 3 seminal papers in the field is described below. The first paper on artificial intelligence in radiology effectively discusses AI advancements in image recognition and task automation but lacks coverage of AI systems' generalizability and ethical concerns. It also overlooks the integration of multimodal data and the need for longitudinal studies to assess long-term clinical effectiveness [33]. The second paper, evaluating an AI system for breast cancer screening, highlights AI's strong performance, surpassing individual radiologists, and features a comprehensive evaluation across varied datasets. However, it has limitations, such as dataset selection biases and the need for further prospective studies to confirm its findings. Additionally, it does not address the long-term impact of AI integration in clinical workflows [22]. The third paper provides valuable statistics on breast cancer incidence and mortality, emphasizing racial disparities and trends, which are critical for public health understanding. While it underscores the role of access to care in addressing these disparities, it may oversimplify the complex socioeconomic factors influencing mortality rates and lacks a deeper exploration of the biological mechanisms contributing to racial differences in outcomes [34]. The keywords in a publication highlight the primary research areas, and the frequency and co-occurrence of these keywords indicate the topics emphasized in that research domain. Our analysis identified “mammography,” “female, ” “artificial intelligence, ” “human, ” and “breast cancer” as the most common keywords. By comparing the most pertinent keywords with top locally and globally cited literature, we observed a significant link between breast cancer and AI technologies, such as Deep Learning and Machine Learning. These keywords are frequently used in the literature and are the most commonly mapped subject areas in our dataset. This finding suggests that researchers in the medical imaging field are primarily focused on addressing challenges in medical imaging through AI techniques, particularly Deep Learning and Machine Learning, to enhance the accuracy of breast cancer screening and prognosis prediction. Additionally, the keyword frequency analysis reflects a shift in research priorities on breast cancer, its diagnosis, and related imaging techniques. Terms like “breast cancer” (1444), “diagnostic imaging” (708), and “digital mammography” (315) emphasise the critical role of imaging in cancer detection. Artificial intelligence (1691), including deep learning (892) and convolutional neural networks (342), is increasingly used for improving diagnostic accuracy (401). The frequent mention of “computer-aided diagnosis” (456) and “algorithms” (432) shows a shift towards AI-driven healthcare solutions. The inclusion of terms like “adult” (556) and “aged” (399) suggests these technologies are applied primarily in adult populations. The relationship between breast cancer and artificial intelligence (AI) tools can be analysed through the cumulative keyword occurrence growth graph from 2000 to 2024. The graph indicates a strong correlation between the keywords “breast cancer,” “breast neoplasm,” “machine learning” (ML), “classification,” “deep learning,” and “machine learning.” This correlation highlights the growing use of AI, especially deep learning, in medical image analysis of breast masses for cancer detection. Consequently, these keywords represent key trending topics in AI for diagnosis and treatment of breast cancer and its patients. The impact of AI technologies on clinical outcomes is multifaceted, encompassing improved diagnostic accuracy, reduced workload for radiologists, enhanced visualization, and the potential for personalized medicine. These advancements can lead to earlier detection and more effective treatment of breast cancer, ultimately improving patient care and outcomes.] The “Make in India” initiative has led to AI advancements like Telerad Tech’s “MammoAssist” and Niramai Health's affordable, non-invasive screening tool, both enhancing early breast cancer detection and radiologist efficiency. Regenerative AI, continuously improving with new data, supports reliable diagnostics across populations. Studies in rural India and sub-Saharan Africa show that mobile health units and digital mammogram analysis have increased screening accuracy and speed, highlighting AI’s impact on breast cancer care in developing regions [35]. AI models in medical imaging diagnosis have evolved to improve diagnostic accuracy, reduce false positive and negative rates, and integrate more effectively into clinical workflows, while also addressing clinician concerns regarding trust and performance. As an example, DeepMind’s algorithms, evaluated in a study, reduced false negatives by 9.4% and false positives by 5.7% but required further validation across diverse populations [35, 36]. Ultimately, the analysis of the social knowledge structure indicates that authors Cassano e, Pesapane f, mann rm, Sechopoulos I, yang y, Nishikawa rm, Lee ci, Elmore JG, Chan h-p, Helvie ma, zheng b, Daniela g, Heidari m, and qiu y have collaborated and published more papers compared to other authors within the dataset. Additionally, the institution collaboration network analysis shows that the University of Pennsylvania (64.8), Karolinska University Hospital (56.6), Karolinska Institute (46.6), and University of California (42.2) have the highest betweenness centrality among all universities in the field of AI for breast cancer diagnosis and treatment. Furthermore, the world map collaboration analysis highlights that developed countries, such as the USA, China, Canada, United Kingdom, Australia, and the Netherlands, are key players in fostering collaborative research on AI for breast cancer diagnosis and treatment through ongoing international partnerships. As an example, an innovative Computer-Aided Detection (CAD) system utilizing deep learning and computer vision techniques was introduced to enhance breast cancer diagnosis by independently identifying, segmenting, and categorizing breast lesions based on pathology. Thorough validation using the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) demonstrated the system’s exceptional performance, with a 99% success rate in detecting and classifying breast masses. The detection accuracy was 98.5%, while the segmentation of breast masses for examination achieved approximately 95.39% performance. In the final classification phase, the system reached an overall accuracy of 99.16% [37]. Some of the most influential collaborations, such as those involving institutions like the University of Pennsylvania, Karolinska University Hospital, and the University of California, have indeed led to breakthroughs in the field. These institutions, with their high betweenness centrality, act as key nodes in the global research network, facilitating knowledge exchange and collaboration across countries. Their work has contributed to major advancements in AI algorithms for early breast cancer detection, personalized treatment plans, and improved diagnostic accuracy. As an example, Magnetic Resonance Imaging (MRI) is key for breast cancer diagnosis, but its multiple volumes can be difficult for radiologists to interpret. To solve this problem, research was proposed by a multimodal framework that processes all MRI data using 3D convolutional neural networks and combines predictions with a late fusion strategy based on Dempster-Shafer theory. The approach improves diagnostic accuracy by integrating multiple modalities, outperforming individual ones. [9] The international collaborations, especially between developed countries like the USA, China, and the UK, have been crucial in accelerating research, allowing for larger datasets, broader clinical trials, and more robust validation processes, ultimately pushing the field toward more impactful innovations. Despite this progress, deep learning (DL) and transfer learning (TL) techniques have yet to be established as primary clinical protocols for breast cancer detection and patient survivability assessment. Thus, global collaboration within the scientific community is essential to undertake the necessary medical device regulations for the integration of deep learning technology in healthcare. Consequently, this systematic bibliometric review serves as a valuable resource for novices aiming to apply Deep Learning and Transfer Learning techniques for breast cancer diagnosis and treatment using various medical imaging modalities. From 2000 to 2024, thematic analysis reveals that significant developments mark the advancement of AI in breast cancer diagnosis and treatment. Key concepts such as Human, AI, Female, Breast neoplasm, Breast cancer, and diagnostic imaging have emerged as crucial elements in the research framework of this field. On the other hand, diseases, medical imaging, computer-aided diagnostics, learning systems, and image segmentation represent emerging or declining themes with weak internal development degrees. They are marginally crucial for developing the given scientific field. According to the Sankey diagram and keywords thematic map, as shown in Fig. 8, we observe that from 2000 to 2021, studies were more focused on applying computer-aided diagnostics for detecting and classifying malignancy and breast cancer. However, during the last two years, the implementation of Artificial intelligence and support of vector machines in the diagnosis and prognosis of breast cancer have paved the way for earlier detection of breast cancer. Our observations indicate that institutions in developed nations rarely take the lead in partnering with counterparts in developing and underdeveloped regions. Collaborations with lower-income countries or institutions with less research funding reveal gaps in the availability and accessibility of AI-based diagnostic tools, shedding light on the disparities within this global research agenda. [38, 39] Developed nations prefer to collaborate with institutions that match or surpass their own in terms of infrastructure and intellectual capability. Therefore, countries with superior infrastructure and economies must engage in collaborative projects with skilled intellectuals and their institutions from developing and underdeveloped nations, leveraging current technologies to create a global AI-driven breast cancer healthcare network. This AI-based network would enable underdeveloped regions to effectively adopt advanced deep learning techniques for breast cancer diagnosis and prognosis. Sharing clinical and imaging data is crucial, yet there is a shortage of annotated, structured, and readily usable data currently limited to a few institutions. Establishing new imaging repositories, such as the Health Data Research Innovation Gateway, is essential to bridge this data gap. These repositories are critical to developing a robust data ecosystem that is necessary for creating innovative algorithms for early detection and treatment response prediction in breast cancer. AI in healthcare, especially in breast cancer diagnosis and treatment, faces several challenges. Effective AI tools rely on high-quality, annotated data, but such data is often scarce, reducing the models' accuracy and generalizability in real-world applications. Integrating AI into existing healthcare systems can also be difficult, as the complexity of the technology and the extensive training required can overwhelm healthcare providers, slowing down adoption. Many AI models function as "black boxes," making it hard for clinicians to understand the reasoning behind the decisions. This lack of transparency can undermine trust, as clinicians may be hesitant to rely on AI without clear explanations. Additionally, AI algorithms may inherit biases from their training data, resulting in unequal performance across diverse patient groups, raising ethical concerns about fairness in breast cancer care [35]. In addition to this, challenges like patient data privacy, model interpretability, and potential algorithmic bias are crucial in clinical AI, especially in oncology. These issues drive research focus and impact public trust and acceptance of AI in healthcare. A thorough examination of these obstacles could offer valuable insights into what is needed to ensure the safe and fair integration of AI in breast cancer diagnosis and treatment. [40] Future research directions may involve the combination of AI with genomics and molecular data to enhance precision medicine, as well as the development of models that can better interpret imaging data with minimal human supervision. However, current barriers include the lack of large, annotated datasets, regulatory hurdles, and concerns over AI's interpretability and ethical use, especially regarding patient privacy and data security. Overcoming these obstacles will require interdisciplinary collaboration, improved data sharing practices, and rigorous validation of AI systems in clinical settings [35]. Limitations Our bibliometric review has several limitations. First, we focused solely on publications available in English. Second, we did not consider electronic preprints from the open-access repository ArXiv, which may have resulted in the omission of several relevant studies on AI and breast cancer diagnosis and prognosis; however, these preprints are not peer-reviewed. Third, our data extraction and analysis were limited to Scopus databases from January 1, 2000, to June 20, 2024, potentially excluding many articles related to AI and breast cancer diagnosis and prognosis published after June 20, 2024. Additionally, it may affect the study’s generalizability. For instance, Scopus may exclude significant studies published in non-indexed or regional journals, particularly those from underrepresented regions. This limitation has implications for the global applicability of the findings and should be interpretated with caution.