The role of artificial intelligence in immune checkpoint inhibitor research: A bibliometric analysis Authors: Ziqi Zhao, Kun Xu, Yizhuo Jiang, Xisheng Xu, Yuliang Liu Published: 2024 ABSTRACT ABSTRACTImmune checkpoint inhibitors (ICIs) are revolutionizing cancer treatment, and Artificial Intelligence (AI) is a key player in this field. A comprehensive analysis of AI’s impact on these inhibitors was lacking, but this study addresses that by analyzing literature for trends and future predictions. It reveals rapid growth and international collaboration. We utilized analytical tools such as CiteSpace, VOSviewer, and PlotDB to analyze 774 documents from the Web of Science Core Collection from 2018 to May 2024, discovering a steady increase in annual publications, with China and the United States leading the way. Sun Yat Sen University and researchers like Ock Chan-young, Zhang Hao, and Newman AM are prominent. The most productive journal is Frontiers in Immunology, while the New England Journal of Medicine has the highest citation rate. The most cited reference is Newman, AM’s 2019 article in Nature Biotechnology. Keywords like “immunotherapy,” “pembrolizumab,” “cancer,” “machine learning,” and “expression” are central to the discourse. Research focuses on the application of inhibitors in non-small cell lung cancer, bioinformatics, and cancer immunotherapy, showing AI’s potential to improve oncology precision medicine. Although AI’s application in ICIs shows promise, significant challenges still demand exploration and resolution. Continued investment in AI research in this context could lead to significant advancements in cancer treatment. Global collaboration is needed to overcome these challenges and fully leverage AI’s potential. This study provides a foundation for future research and interdisciplinary collaboration in this critical field. Introduction Immune checkpoint inhibitors (ICIs) are a class of drugs used to treat certain types of cancer. They work by blocking immune checkpoints to enhance the immune system’s attack on cancer cells. Immune checkpoints are “brake” mechanisms within the immune system that usually prevent it from attacking our own cells. However, cancer cells sometimes exploit these checkpoints to evade immune surveillance and attack. ICIs include PD-1 (Programmed Death Protein 1) inhibitors and CTLA-4 (Cytotoxic T-Lymphocyte-Associated Protein 4) inhibitors, among others. PD-1 inhibitors, such as nivolumab (Opdivo) and pembrolizumab (Keytruda), enhance T-cell attacks on cancer cells by blocking PD-1.1–4 CTLA-4 inhibitors, such as ipilimumab (Yervoy),5 enhance the immune system’s attack on tumors by blocking CTLA-4. According to the latest research progress, the application of ICIs has evolved from monotherapy to combination therapy,6 including combinations with chemotherapy, radiotherapy,7 targeted therapy,8,9 and emerging immunotherapies. This multimodal treatment strategy aims to improve clinical efficacy through synergistic effects and more comprehensively target tumors. For instance, the combination of PD-1/PD-L1 inhibitors with radiotherapy can increase tumor immunogenicity, thereby enhancing the effectiveness of immunotherapy.10,11 Additionally, the new generation of ICIs, such as antibodies targeting LAG-3 and TIGIT,12 also show promise in clinical research. In the medical field, Artificial Intelligence (AI) is rapidly becoming a significant driving force for industry innovation. AI technologies have been widely applied in various aspects, including auxiliary diagnosis,13 clinical decision support,14 drug development,15 personalized medicine,16 patient monitoring,17 surgical assistance robots,18 epidemiological monitoring, public health decision-making,19 medical education and training,16 and optimization of medical resources.20 Recently, AI and deep learning technologies have made tremendous progress in the field of cancer diagnosis and treatment, particularly in the research and application of ICIs. AI technologies, through automated feature extraction,21 image recognition,22 and big data analysis,23 have opened up new horizons for the early detection, precise classification, accurate grading, treatment selection, and prognosis assessment of cancer. In the field of ICI research, the application of AI technology has covered several important aspects, including predicting patient responses to treatment,24 optimizing therapeutic combination strategies,25 managing immune-related adverse events (irAEs),26 drug discovery,27 and tumor microenvironment analysis.28 AI technology aids scientists in predicting patient responses to ICIs by analyzing biomarkers such as tumor mutational burden (TMB) and PD-L1 expression.26 Additionally, the application of AI in digital pathology provides researchers with new tools for in-depth analysis of the tumor microenvironment, including the distribution, function, and interactions of immune cells, which helps to deepen the understanding of the mechanisms of action of ICIs and develop new biomarkers to predict patient responses.29 Despite the broad prospects for the application of AI technology in ICI research, it also faces some challenges and limitations. These include algorithm validation, interpretability issues, the demand for computing resources, and the acceptance of new technologies by medical professionals. AI models are often seen as “black boxes,” and their decision-making opacity limits the trust of medical professionals.30,31 Moreover, obtaining high-quality, diverse datasets is particularly challenging in some specialized fields. Therefore, future research needs to develop more effective data collection and annotation methods and use techniques such as synthetic data and federated learning to overcome data scarcity issues. As AI technology becomes increasingly widespread in the medical field, ethical and privacy issues also become increasingly important. Future research needs to explore how to use patients’ health data to train AI models reasonably while protecting individual privacy. In addition, interdisciplinary collaboration and educational reform are crucial for promoting the application of AI technology in the medical field, requiring the joint participation of experts in computer science, medicine, ethics, and other fields, and updating medical education to adapt to the new requirements of digital medicine. Given the swift pace of progress in both ICIs and AI, it is challenging for researchers to stay abreast of the latest findings and anticipate future trends. Bibliometric analysis, a research method that uses quantitative analysis of publications to study the development of scientific fields, offers a structured approach to understanding the evolution of research areas.32 By examining publication trends, citation patterns, and author collaborations, bibliometric studies can provide insights into the growth, impact, and direction of scientific disciplines. The literature suggests AI’s promise in boosting the effectiveness of ICIs, but a thorough bibliometric study of this area is lacking. This research seeks to address this gap by using bibliometric methods to scrutinize the literature on AI’s role in this context. Our goal is to pinpoint major research topics, notable publications, and future trends. This in-depth analysis will help us better grasp how AI can optimize cancer immunotherapy and enhance patient outcomes. Materials and methods Searching strategy and data collection This study employed a cross-sectional research design and collected data through the Science Citation Index Expanded (SCI-E) and the Social Sciences Citation Index (SSCI) via the Web of Science (WOS). Our literature search was conducted up to May 29, 2024. To ensure data consistency and facilitate subsequent analysis, we uniformly acquired all documents in “plain text” format. During the data screening phase, we deliberately excluded articles from years with publication counts of less than ten. This strategy aims to maintain high standards of data quality and further enhance the accuracy of our analysis results, based on the principle that limited datasets may skew statistical outcomes and visualizations. Bibliometrics is a powerful methodological tool that can reveal the development trends of a specific discipline or research field over a certain period. To ensure the accuracy and authority of research results, selecting a comprehensive and representative database is essential. The Web of Science (WoS) is a multidisciplinary database that includes numerous high-impact scientific journals and prestigious indexes. Compared to Scopus or MEDLINE/PubMed, WoS offers more extensive information for bibliometric analysis.33,34 The Web of Science Core Collection (WoSCC) is highly regarded for its comprehensiveness and authority, covering numerous prestigious academic journals, offering resource for bibliometric studies.35 However, not all sub-databases within WoSCC are suitable for bibliometric analysis. Based on prior experience, the SCI-E database is favored for its broad applicability and acceptance, while the SSCI database is known for its high data quality and inclusion of top-tier journals. The SSCI also offers extensive functionality for literature searches and advanced analysis.36,37 Therefore, we have chosen to use the SCI-E and SSCI databases as our data sources to ensure accurate and representative findings for our research. The search formula is as follows: TS=(“immune checkpoint inhibitors” OR “immune checkpoint blockade” OR “immunological checkpoint inhibitor” OR “immune checkpoint inhibitor” OR “immunological checkpoint inhibitors” OR “immuno-checkpoint inhibitors” OR “immune checkpoint blockers” OR“AntiCTLA-4” OR “Anti-PD-1” OR “Anti-PD-L1” OR “Ipilimumab” OR “Tremelimumab” OR “Pembrolizumab” OR “Atezolizumab” OR ICIs) AND TS = (((automated OR intelligent) NEAR/1 (classification OR diagnosis OR segment* OR detect*)) OR “artificial intelligence” OR “deep learning” OR “convolutional neural network*” OR “machine learning” OR “CNNs” OR “artificial neural network*” OR “computer-aided” OR “Bayes* network*” OR “supervised learning” OR “unsupervised clustering” OR “computer-assisted” OR (deep NEAR/1 network*) OR “ensemble learning”) When constructing the search query, we first conducted a preliminary search to assess the keywords that emerged in the fields of AI and ICIs. Subsequently, we carefully selected and categorized these keywords to ensure the accuracy and comprehensiveness of the search query. From the retrieved relevant publications, we collected information such as titles, abstracts, keywords, authors, institutions, countries/regions, and references, with the stipulation that the publications were written in English and limited to original articles or reviews. Figure 1 outlines our study’s literature search and selection process, including exclusion criteria. To ensure the precision and effectiveness of the screening results, we have adopted a double-blind screening process. This process is carried out by two independent researchers who are responsible for meticulously organizing and thoroughly discussing any questionable literature encountered during the screening process to jointly decide whether these documents are suitable for inclusion in our final research scope. In cases of significant disagreement, we submit these disputes to our team’s guiding professors for their final decision. Figure 1.Literature selection flowchart for bibliometric analysis on ICIs and AI. Literature selection flowchart for bibliometric analysis on ICIs and AI. Data analysis and visualization We input the collected data into tools such as CiteSpace, VOSviewer, and PlotDB, all of which are renowned for their exceptional performance in bibliometric analysis and visualization. CiteSpace and VOSviewer are highly regarded in academia, with VOSviewer particularly known for its intuitive and straightforward user interface, focusing on the visualization of bibliometric networks, which gives it a clear advantage in creating clear and easily understandable charts. While CiteSpace offers a wealth of parameter settings for customizing network visualization, its interface is relatively complex, which may impact the user experience.38 CiteSpace excels in handling large datasets and performing complex analyses, supporting advanced features such as keyword co-occurrence networks, co-clustering analysis, and burst term detection, making it a powerful tool for exploring research trends and hotspots.39 In contrast, VOSviewer is more suitable for dealing with smaller datasets, and its mapping functions focus on clearly presenting the connections between data.40 PlotDB, as an open platform, offers a rich selection of chart options. By leveraging the unique strengths of these tools, we conducted a comprehensive analysis of key elements such as countries, institutions, authors, references, and keywords, integrating existing research findings. Our aim is to use these tools to draw a comprehensive knowledge map that reveals research trends, key elements, and their interrelationships. This analysis not only clarifies the current state of research but also provides valuable guidance and insights for future research directions. In this study, we utilized the widely recognized academic tool CiteSpace 6.3 R1 (64-bit)41 to conduct an in-depth knowledge visualization analysis on the application of AI in the field of ICIs. The analysis encompassed multiple dimensions, including countries, journals, keywords, and literature, to explore their centrality, co-occurrence relationships, time series, and burst detection, in order to reveal the knowledge base of the field and its evolution over time. CiteSpace is not only capable of depicting a dual-map visualization of journal connection networks based on citation and co-citation relationships but can also identify the surge of research topics by examining the significant increase in citation frequency within a specific time window, which typically indicates that emerging research topics are receiving rapid attention from the academic community. Through this analysis, we are able to gain insights into the key elements and development trends within the field, providing strong support for understanding the knowledge structure of the domain. VOSviewer is a popular bibliometric tool that visualizes the connections within scientific literature to map knowledge domains.42 We utilized VOSviewer for co-occurrence analysis of countries, institutions, keywords, and references. It employs VOS mapping technology to achieve this by constructing a distance map based on a similarity matrix. VOSviewer supports the visualization of knowledge maps for large-scale literature data and has been widely used in bibliometric analysis research. The visualization map is composed of nodes and links. The size of the node circles is proportional to their frequency of occurrence. The links between nodes demonstrate the strength and relevance of their connections, with the thickness and length of the links representing the tightness and closeness of these connections, respectively. In cluster analysis, the color variation of circles intuitively indicates their category affiliation. To visualize certain raw data, such as creating statistical charts for author centrality, we utilized the PlotDB online platform. Results Chronological publication analysis Between January 1, 2018, and May 29, 2024, we found 774 publications on the intersection of ICIs and AI. Original research made up the majority, with 619 articles (79.97%), while reviews totaled 155 (20.03%). The field has experienced substantial growth, with only 13 publications in 2018, but a dramatic rise to 216 by 2023—an increase of 1561.5% (as shown in Figure 2). All publications received a total of 12,641 citations, and 11,959 when excluding self-citations. The average citations per publication were 16.33, and the H-index was 48, indicating the research’s quality and influence. Figure 2.Trends in the number of publications and citations for ICIs and AI (2018–2024). Trends in the number of publications and citations for ICIs and AI (2018–2024). Distribution of countries/regions, institutions and funding agencies Fifty countries or regions have engaged in AI research for ICIs. Table 1 ranks the top 10 contributors by publication volume and centrality. China and the U.S. lead with over 80% of the total publications. Germany, Italy, and South Korea also show strong participation. The centrality index measures a country’s research impact and prominence in the field. Figure 3a, visualized with Citespace, shows each country’s centrality, with purple circles highlighting those above 0.1. The top five in centrality are Portugal (0.89), Egypt (0.86), Belgium (0.83), Austria (0.58), and Spain (0.41). This suggests that high publication volume does not equate to centrality, indicating a need for these countries to enhance global research collaboration.Table 1.Top 10 countries by volume and centrality of publications.RankCountryDocumentsCitationsTotal Link StrengthRankCountryCentrality1China4164243761Portugal0.892USA21166981982Egypt0.863Germany601264933Belgium0.834Italy511206834Austria0.585South Korea42490225Spain0.416France371237606Denmark0.397United Kongdom36790917Netherlands0.348Japan32275218Italy0.299Netherlands21623539Germany0.2810Canada201062310USA0.22Figure 3. Global distribution of AI research in ICIs: collaboration and centrality analysis. (a) National centrality node map. (b) Map of cooperation between countries. (c) Institution centrality circular diagram. (d) Institution’s’s document output and average citation ratio chart. (e) Institution cooperation contact graph.Figure 3. Continued.Figure 3.Continued. Table 1.Top 10 countries by volume and centrality of publications.RankCountryDocumentsCitationsTotal Link StrengthRankCountryCentrality1China4164243761Portugal0.892USA21166981982Egypt0.863Germany601264933Belgium0.834Italy511206834Austria0.585South Korea42490225Spain0.416France371237606Denmark0.397United Kongdom36790917Netherlands0.348Japan32275218Italy0.299Netherlands21623539Germany0.2810Canada201062310USA0.22 Top 10 countries by volume and centrality of publications. Global distribution of AI research in ICIs: collaboration and centrality analysis. (a) National centrality node map. (b) Map of cooperation between countries. (c) Institution centrality circular diagram. (d) Institution’s’s document output and average citation ratio chart. (e) Institution cooperation contact graph. Continued. Continued. Furthermore, Figure 3b uses VOSviewer to display the cooperation between countries. It is clear from the figure that the United States occupies the top position in terms of link strength, and the cooperation between China and the United States is particularly close. This indicates that the two countries have an important synergistic effect in this research field, jointly promoting the advancement of AI and immune checkpoint inhibitor research. This study includes 275 research institutions, with Sun Yat-sen University in China publishing the most, 44 times, followed by Peking Union Medical College and Shanghai Jiao Tong University with 26 each. Table 2 and Figures 3c,d show that Chinese institutions make up 80% of the top 10 in publication volume, highlighting China’s significant contribution to AI and ICIs research.Table 2.Top Ten institutions by publication volume.RankInstitutionCountCentralityCitationAverage CitationCountry1Sun Yat Sen University440.0154412.36China2Peking Union Medical College260.2346517.88China3Shanghai Jiao Tong University26026110.04China4Southern Medical University - China250.0726110.44China5Institut National de la Sante et de la Recherche Medicale (Inserm)240.01102242.58Franch6Fudan University2401807.5China7Chinese Academy of Sciences240.1451121.29China8State Key Lab Oncology South China240.0639416.42China9Sichuan University2302299.96China10Harvard University220.1338617.55USA Table 2.Top Ten institutions by publication volume.RankInstitutionCountCentralityCitationAverage CitationCountry1Sun Yat Sen University440.0154412.36China2Peking Union Medical College260.2346517.88China3Shanghai Jiao Tong University26026110.04China4Southern Medical University - China250.0726110.44China5Institut National de la Sante et de la Recherche Medicale (Inserm)240.01102242.58Franch6Fudan University2401807.5China7Chinese Academy of Sciences240.1451121.29China8State Key Lab Oncology South China240.0639416.42China9Sichuan University2302299.96China10Harvard University220.1338617.55USA Top Ten institutions by publication volume. While the Institut National de la Santé et de la Recherche Médicale (Inserm) has the highest total and average citations, its centrality is less notable. In contrast, Peking Union Medical College has the highest centrality score of 0.23, signifying its pivotal and influential position in the field. Further analysis of the 29 most prolific institutions, using the VOSviewer map in Figure 3e, shows two main groups: Chinese institutions on the left and American on the right. Chinese institutions are closely linked but interact less with American ones. Sun Yat-sen University heads a green cluster of southeast Chinese research and universities. The red cluster connects Chinese and American institutions, with Peking Union Medical College and Stanford University as key nodes. The blue cluster, consisting mainly of U.S., South Korean, and French institutions and led by the U.S., has limited internal cooperation and collaboration with others, suggesting opportunities for enhanced international partnerships. Funding agencies are crucial for driving scientific research and academic publishing. Table 3 lists the top 10 agencies that sponsored over 10 articles in this field. Half of these are Chinese, and three are American. The National Natural Science Foundation of China, the U.S. Health Human Services, and the National Institutes of Health are the top three funders, showing the U.S.‘s ongoing leadership and China’s swift emergence in this scientific area, supported by its robust economy and research investment.Table 3.10 major sponsoring institutions with more than 10 sponsored articles.RankAwardsCountCountryH-index1National Natural Science Foundation of China NSFC176China242United States Department of Health Human Services63USA233National Institutes of Health NIH USA60USA204NIH National Cancer Institute NCI27USA125National Key R D Program of China20China106China Postdoctoral Science Foundation17China67National Research Foundation of Korea17Korea78Bristol Myers SQUIBB15China119National Natural Science Foundation of Guangdong Province13China610Japan Society for the Promotion of Science11Japan5 Table 3.10 major sponsoring institutions with more than 10 sponsored articles.RankAwardsCountCountryH-index1National Natural Science Foundation of China NSFC176China242United States Department of Health Human Services63USA233National Institutes of Health NIH USA60USA204NIH National Cancer Institute NCI27USA125National Key R D Program of China20China106China Postdoctoral Science Foundation17China67National Research Foundation of Korea17Korea78Bristol Myers SQUIBB15China119National Natural Science Foundation of Guangdong Province13China610Japan Society for the Promotion of Science11Japan5 10 major sponsoring institutions with more than 10 sponsored articles. Analysis of the active authors and co-cited authors In this study’s literature review, we identified 6,686 authors and 22,099 co-cited authors, averaging 8.64 authors per paper. Table 4 and Figure 4a,b show the top 4 authors by publication count and co-citation rate. The most prolific author, Ock, Chan-young, has only 8 publications, indicating no dominant figures in terms of output. This suggests that AI’s use in ICIs research is still fragmented, lacking a consolidated research community. Figure 4.Author contributions and Co-citations in AI and Immune checkpoint inhibitor research. (a) Author citation ranking circular diagram. (b) Co-cited Author ranking top 10 circular diagram.Table 4.Top 10 authors by publication count and total cited frequency.RankAuthorDocumentsCitationsCountryCo-cited AuthorCitationsTotal Link StrengthCountry1Ock,Chan-young8160South KoreaReck, M1702452Germany2Zhang, Hao871ChinaHerbst, RS1292092USA3Li, Yan7135ChinaRobert, C1141342France4Tian, Jie7330ChinaNewman, AM1051352USA5Ammari, Samy6776FranceHellmann, MD1021663USA6Dercle, Laurent6819IranYoshihara, K981331Japan7Chen, Wei553ChinaJiang, P971330China8Kim, Seokhwi5114South KoreaHänzelmann, S921239Germany9Lu, Shun5138ChinaSiegel, RL901037USA10Marabelle, Aurelien5809FranceBorghaei, H851309USA Author contributions and Co-citations in AI and Immune checkpoint inhibitor research. (a) Author citation ranking circular diagram. (b) Co-cited Author ranking top 10 circular diagram. Table 4.Top 10 authors by publication count and total cited frequency.RankAuthorDocumentsCitationsCountryCo-cited AuthorCitationsTotal Link StrengthCountry1Ock,Chan-young8160South KoreaReck, M1702452Germany2Zhang, Hao871ChinaHerbst, RS1292092USA3Li, Yan7135ChinaRobert, C1141342France4Tian, Jie7330ChinaNewman, AM1051352USA5Ammari, Samy6776FranceHellmann, MD1021663USA6Dercle, Laurent6819IranYoshihara, K981331Japan7Chen, Wei553ChinaJiang, P971330China8Kim, Seokhwi5114South KoreaHänzelmann, S921239Germany9Lu, Shun5138ChinaSiegel, RL901037USA10Marabelle, Aurelien5809FranceBorghaei, H851309USA Top 10 authors by publication count and total cited frequency. Ock, Chan-young’s research focuses on the application of AI technology in the analysis of tumor-infiltrating lymphocytes (TIL) and the role of radiomics analysis in predicting TIL enrichment and the inflammatory immune phenotype (IIP) in the prognostic prediction of immune checkpoint inhibitor therapy.43,44 One of his highly cited articles combines clinical trials, safety assessments, efficacy analysis, and AI technology, providing valuable insights and data support for neoadjuvant immunotherapy in head and neck squamous cell carcinoma.45 Reck, M, as the author with the most co-citations, focuses on various treatment methods for non-small cell lung cancer, including research on immunotherapy, chemotherapy, targeted therapy, and biomarkers. Reck, M’s contributions to the field of ICIs include a comparative study of nivolumab (a PD-1 inhibitor) with standard chemotherapy drug docetaxel in patients with advanced squamous cell NSCLC.46 This study confirmed the significant advantages of nivolumab in improving overall survival, response rate, and progression-free survival, regardless of PD-L1 expression levels. He also participated in research on the efficacy of pembrolizumab (another PD-1 inhibitor) combined with chemotherapy drugs pemetrexed and platinum-containing drugs in patients with metastatic non-squamous NSCLC, demonstrating that the combination therapy can significantly extend the overall survival and progression-free survival of patients.47 In addition, Reck, M has explored the correlation between tumor mutational burden and response to immunotherapy,48 providing preliminary evidence that tumor mutational burden may increase the benefit of immunotherapy, especially in patients with higher tumor mutational burden, where the efficacy of nivolumab may be more pronounced. Analysis of top journals and co-cited journals This study examines AI applications in ICIs, finding related publications spread across 275 journals and 4,039 co-cited ones. Table 5 lists the top 10 most influential journals, with details on publication volume, citations, impact factors, and JCR categories, reflecting their academic prestige. A majority, 60%, are in the Q1 category, showing their leading role in academia and a focus on the interdisciplinary research between medical oncology and molecular biology.Table 5.The top ten influential journals in terms of publication and citation, along with their publication volume, citation counts, impact factors, and JCR categories.RankJournalDocumentsCitationsIFJCRCo-cited JournalCitationsIFJCR1Frontiers in Immunology665347.3Q1New England Journal of Medicine1606158.5Q12Cancers543815.2Q1Journal of Clinical Oncology135245.4Q13Frontiers in Oncology423144.7Q2Clinical Cancer Research109511.5Q14Scientific Reports222264.6Q2Nature105764.8Q15Journal for Immunotherapy of Cancer1939310.9Q1Annals of Oncology89350.5Q16Frontiers in Genetics13583.7Q1Cell87064.5Q17International Journal of Molecular Sciences13845.6Q1/Q2Nature Communications83716.6Q18Clinical Cancer Research1214811.5Q1Nature Methods83048Q19BMC Cancer11533.8Q2Lancet Oncology82351.1Q110Nature Communications1156016.6Q1Science74756.9Q1 Table 5.The top ten influential journals in terms of publication and citation, along with their publication volume, citation counts, impact factors, and JCR categories.RankJournalDocumentsCitationsIFJCRCo-cited JournalCitationsIFJCR1Frontiers in Immunology665347.3Q1New England Journal of Medicine1606158.5Q12Cancers543815.2Q1Journal of Clinical Oncology135245.4Q13Frontiers in Oncology423144.7Q2Clinical Cancer Research109511.5Q14Scientific Reports222264.6Q2Nature105764.8Q15Journal for Immunotherapy of Cancer1939310.9Q1Annals of Oncology89350.5Q16Frontiers in Genetics13583.7Q1Cell87064.5Q17International Journal of Molecular Sciences13845.6Q1/Q2Nature Communications83716.6Q18Clinical Cancer Research1214811.5Q1Nature Methods83048Q19BMC Cancer11533.8Q2Lancet Oncology82351.1Q110Nature Communications1156016.6Q1Science74756.9Q1 The top ten influential journals in terms of publication and citation, along with their publication volume, citation counts, impact factors, and JCR categories. Frontiers in Immunology publishes the most articles in this field with 66, Cancers follows with 54, and Frontiers in Oncology with 42. The New England Journal of Medicine has the highest co-citation count at 1,606, while the Journal of Clinical Oncology and Clinical Cancer Research have 1,352 and 1,095 respectively. Clinical Cancer Research and Nature Communications, despite not being top-ranked, are notably influential, ranking 3rd and 7th in co-citation, indicating their importance in AI and ICIs research. Figure 5‘s dual-layer map illustrates how journals cite and reference each other, uncovering four key citation pathways: mutual citations among molecular biology, immunology, and genetics; cross-referencing between these fields and health, nursing, and medicine; links between medicine, clinical science, and genetics; and interactions among medicine, clinical science, and health-related disciplines. This analysis shows that citing papers are mainly in molecular biology, immunology, medicine, and clinical science, with some citations in physics, materials science, and other fields. Cited papers are mostly in health, medicine, and genetics, with additional mentions in environmental science, toxicology, and other disciplines. The visualization highlights interdisciplinary connections and the trend toward cross-disciplinary research, offering insights into the academic influence and knowledge flow in this area. Figure 5.Dual-graph overlay of citing and cited journals in AI and Immune checkpoint inhibitor research. Dual-graph overlay of citing and cited journals in AI and Immune checkpoint inhibitor research. Keywords concentration zones Keywords are pivotal for grasping the essence and trends in a research domain. Our in-depth analysis identified the top 20 keywords, such as “expression,” “pembrolizumab,” “cancer,” “nivolumab,” “survival,” “open-label,” “immunotherapy,” “therapy,” “chemotherapy,” and “PD-1 blockade,” as listed in Table 6. VOSviewer’s co-occurrence analysis uncovered 2,692 keywords, with 58 appearing frequently, forming four main clusters (as seen in Figure 6a). The red cluster, emphasizing “immunotherapy,” “expression,” and “cancer,” concentrates on cancer immunotherapy and biomarker discovery. The green cluster, with “machine learning,” “immune checkpoint inhibitors,” and “artificial intelligence,” discusses cancer diagnosis, treatment, and prognosis, including AI applications in medicine. The blue cluster, featuring “immunotherapy drugs” and “chemotherapy,” evaluates the combination treatment’s efficacy and safety, seeking optimal plans through trials and analysis. The yellow cluster, though small with “Ipilimumab” and “melanoma,” is closely linked to other clusters, indicating the interconnected nature of the research field. Ipilimumab, as an immune checkpoint inhibitor blocking CTLA-4, is of great significance for the treatment of advanced melanoma.49 Melanoma, as a highly malignant skin tumor, has a significant impact on the prognosis of advanced-stage patients.50 Figure 6.Keyword Co-occurrence and cluster timeline analysis in AI and Immune checkpoint inhibitor research. (a) Keyword Co-occurrence graph. (b) Keyword cluster timeline chart.Table 6.Top Ten high-frequency keywords.RankKeywordOccurrencesTotal Link Strength1Immunotherapy28611012Pembrolizumab1396483Cancer1354404Machine learning1284825Expression1274896Nivolumab1115587Survival1074358Open-label985049Immune checkpoint inhibitors9138710Tumor microenvironment89286 Keyword Co-occurrence and cluster timeline analysis in AI and Immune checkpoint inhibitor research. (a) Keyword Co-occurrence graph. (b) Keyword cluster timeline chart. Table 6.Top Ten high-frequency keywords.RankKeywordOccurrencesTotal Link Strength1Immunotherapy28611012Pembrolizumab1396483Cancer1354404Machine learning1284825Expression1274896Nivolumab1115587Survival1074358Open-label985049Immune checkpoint inhibitors9138710Tumor microenvironment89286 Top Ten high-frequency keywords. To explore the main themes in AI applications for ICIs (ICIs), we used Citespace for a timeline analysis, depicted in Figure 6b. This method enabled us to follow the development of each keyword cluster chronologically, offering a detailed view of research trends. The visualization shows 10 key clusters, with their horizontal placement indicating the period of their prominence. Node size represents keyword frequency, and colored lines show connections between clusters. The figure clearly shows that foundational keywords like “therapy,” “chemotherapy,” “immune checkpoint inhibitors,” and “expression” have been consistently present since the field’s beginning, as indicated by their large node size, reflecting significant research impact. As time progresses, new research interests have emerged, particularly in areas like computed tomography (CT) and prostate cancer. These new focal points illustrate the field’s evolving nature and its expanding research scope. Knowledge base development and intensive growth In this study, we collected a total of 774 papers, among which 46 received more than 50 citations. Table 7 displays the top 10 most cited papers. Leading the list is the 2019 study by Newman, AM, and his team, who introduced a computational framework called CIBERSORTx.51 This framework is capable of precisely inferring the abundance of cell types and their specific gene expression from tissue RNA profiles. The paper has garnered a total of 1901 citations. Following closely is the 2018 research by Sun, R, and colleagues, who developed and validated a radiomics-based imaging biomarker for CD8 cells.52 This biomarker is used to predict the response and clinical outcomes of patients with advanced solid tumors to single-agent immunotherapy targeting PD-1 or PD-L1.Table 7.Top Ten most cited articles.RankPaperJournalFirst AuthorYearTotal Citations1Determining cell type abundance and expression from bulk tissues with digital cytometryNature BiotechnologyNewman, AM201919012Aradiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort studyLancet OncologySun, R20187223A structured tumor-immune microenvironment in triple negative breast cancer revealed by multiplexed Ion Beam imagingCellKeren, L20185654Lung cancer LDCT screening and mortality reduction-evidence, pitfalls and future perspectivesNature Reviews Clinical OncologyOudkerk, M20212225Use of Immunotherapy With Programmed Cell Death 1 vs Programmed Cell Death Ligand 1 Inhibitors in Patients With Cancer A Systematic Review and Meta-analysisJAMA OncologyDuan, JC20202076Machine learning-based integration develops an immune-derived lncRNA signature for improving outcomes in colorectal cancerNature CommunicationsLiu, ZQ20222017Changes in CT Radiomic Features Associated with Lymphocyte Distribution Predict Overall Survival and Response to Immunotherapy in Non-Small Cell Lung CanerCancer Immunology ResearchKhorrami, M20201638Intestinal microbiota signatures of clinial response and immune-related adverse events in melanoma patients treated with anti-PD-1Nature MedicineMcCulloch, JA20221569Identification of neoantigens for individualized therapeutic cancer vaccinesNature Reviews Drug DiscoveryLang, F202215510Molecular classification and therapeutic targets in extrahepatic cholangiocarcinomaJournal of HepatologyMontal, R2020155 Table 7.Top Ten most cited articles.RankPaperJournalFirst AuthorYearTotal Citations1Determining cell type abundance and expression from bulk tissues with digital cytometryNature BiotechnologyNewman, AM201919012Aradiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort studyLancet OncologySun, R20187223A structured tumor-immune microenvironment in triple negative breast cancer revealed by multiplexed Ion Beam imagingCellKeren, L20185654Lung cancer LDCT screening and mortality reduction-evidence, pitfalls and future perspectivesNature Reviews Clinical OncologyOudkerk, M20212225Use of Immunotherapy With Programmed Cell Death 1 vs Programmed Cell Death Ligand 1 Inhibitors in Patients With Cancer A Systematic Review and Meta-analysisJAMA OncologyDuan, JC20202076Machine learning-based integration develops an immune-derived lncRNA signature for improving outcomes in colorectal cancerNature CommunicationsLiu, ZQ20222017Changes in CT Radiomic Features Associated with Lymphocyte Distribution Predict Overall Survival and Response to Immunotherapy in Non-Small Cell Lung CanerCancer Immunology ResearchKhorrami, M20201638Intestinal microbiota signatures of clinial response and immune-related adverse events in melanoma patients treated with anti-PD-1Nature MedicineMcCulloch, JA20221569Identification of neoantigens for individualized therapeutic cancer vaccinesNature Reviews Drug DiscoveryLang, F202215510Molecular classification and therapeutic targets in extrahepatic cholangiocarcinomaJournal of HepatologyMontal, R2020155 Top Ten most cited articles. This study compiled papers that cited a total of 30,361 references, highlighting the depth of research in this field. Notably, 53 documents were frequently cited at least 30 times, making them a focus for our further analysis. Using the VOSviewer tool, we conducted a co-citation analysis and visualization of these highly cited documents (see Figure 7a), revealing three main research clusters. The red cluster focuses on the application of ICIs in the treatment of non-small cell lung cancer (NSCLC), delving into drugs targeting the PD-1/PD-L1 and CTLA-4 pathways.46,53 The green cluster concentrates on bioinformatics and systems biology, especially on gene expression data analysis,54 cancer genomics,55,56 and the discovery and validation of biomarkers for response to immunotherapy. This cluster also includes methodological approaches such as Weighted Gene Co-expression Network Analysis (WGCNA).57 The blue cluster of articles collectively addresses cancer immunotherapy, particularly the application of PD-1 and its ligand PD-L1 ICIs in the treatment of various cancers. The research emphasizes understanding how the tumor immune microenvironment,58 tumor mutational burden (TMB),59 PD-L1 expression, Epstein-Barr virus (EBV) status, microsatellite instability (MSI), and other genomic and transcriptomic features affect patients’ responses to PD-1 blockade therapy.Figure 7. Highly cited literature Co-citation and clustering timeline in AI and Immune checkpoint inhibitor research. (a) Highly cited literature Co-citation cluster analysis. (b) Highly cited literature clustering timeline chart. (c) Burst literature graph.Figure 7.Continued. Highly cited literature Co-citation and clustering timeline in AI and Immune checkpoint inhibitor research. (a) Highly cited literature Co-citation cluster analysis. (b) Highly cited literature clustering timeline chart. (c) Burst literature graph. Continued. Co-citation analysis reveals the historical development of research within a field. Figure 7b identifies 10 key clusters that show how research interests have evolved. Early on, the focus was on melanoma and cancer genomics, with pembrolizumab as a key treatment. Later, around 2015, there was a notable expansion into the gut microbiome and hepatocellular carcinoma, highlighting the emerging connection between gut health and cancer treatment. By 2017, the spotlight moved to pancreatic cancer, non-small cell lung cancer, and network meta-analysis, showing an increased application of ICIs across different cancers. The continued research on hepatocellular carcinoma underscores its significance as a current hot topic. Figure 7c shows a notable rise in reference citations, divided into three phases. The first, starting in 2018, concentrated on tumor immunotherapy, especially for non-small cell lung cancer (NSCLC) and melanoma, with a focus on PD-1 and PD-L1 inhibitors. Research delved into the efficacy of these inhibitors, biomarker identification, and the immune microenvironment, offering insights into immunotherapy mechanisms and strategies. Entering the second phase in 2019, there was significant advancement in understanding tumor immunotherapy complexities, new biomarker discovery, and personalized treatment development. This phase expanded genomic analyses, linking neoantigen load to treatment response and exploring RNA-level information for neoantigen prediction. The third phase from 2020 to 2022 aimed to improve cancer immunotherapy effectiveness, precision, and patient prognosis. It emphasized optimizing standards, clinical trial innovation, and a deeper dive into the tumor immune microenvironment. A 2022 high-impact paper on cancer genomics, analyzing somatic variations, also emerged, indicating ongoing research evolution toward more effective, personalized cancer treatments. The paper introduced Maftools,60 an R Bioconductor package that provides an efficient and comprehensive tool for the analysis and visualization of somatic variation data in cancer genomics research, which is expected to accelerate progress in this field. Discussion In a short span of time, AI research has made leaps, applying its insights broadly, especially in medicine. Unlike standard reviews, bibliometric analysis uses tools to deeply analyze literature and visually reveal research trends and future hotspots. This study is the first to use bibliometric methods to summarize how AI is applied in ICIs. Using VOSviewer and CiteSpace, we’ve shown the field’s development and future potential, giving new insights to scholars. Looking at the past seven years, we see a clear growth in AI’s application in ICIs, both in the number of studies and their quality, as shown by rising publication and citation counts. The literature indicates a consistent yearly increase in publications. Metrics like average citations per article and total yearly citations are closely tied to publication numbers and time. Recent literature from 2023 and 2024 might not yet reflect the full impact of the work, but the data suggests ongoing growth and potential in this field. China leads in total publications (Table 1), with its global contributions rising annually, reflecting a strong focus on this research area. However, despite leading in publication numbers, China is not in the top 10 for centrality, and the U.S., with the second-highest volume, ranks only 10th in centrality. This suggests a need for research with greater impact. Additionally, the international cooperation network (Figure 3b) shows the U.S. has the most connections and the strongest links overall, highlighting its active role in global scientific collaboration. In the AI and ICIs research, the top three journals by article volume are “Frontiers in Immunology” (IF 7.3, Q1), “Cancers” (IF 5.2, Q1), and “Frontiers in Oncology” (IF 4.7, Q2). These journals’ impact factors, JCR categories, and total citations are key measures of their academic standing. Notably, “Nature Communications,” despite lower publication volume, has a high total citation count, reflecting the high quality and significant impact of its articles. Observing the citation patterns, citing and cited journals are largely concentrated in two specific areas with frequent mutual citations but fewer connections with other fields. This indicates that the field could benefit from more interdisciplinary cooperation to foster knowledge and technology integration and enhance the field’s overall development. Research collaboration’s impact is substantial. China has a strong presence, with 8 of the top 10 institutions in this field (Table 2), highlighting its growing prominence and evolution into a major research hub. The total citation count is a key measure of publication quality. While Inserm is fifth in publication volume, it leads in total citations at 1,022, followed by Sun Yat Sen University (544) and the Chinese Academy of Sciences (511). Top-ranking institutions, aside from Chinese ones, are from developed countries, showing their emphasis on AI’s medical applications. China should sustain close ties with these nations, leveraging technological exchanges and mutual learning to boost its research influence and competitiveness. Further coauthor analysis (see Table 4) reveals an interesting phenomenon: even among the authors with the highest number of publications, no one has published more than 10 articles, indicating that the research field has not yet formed a mature and systematic research system. Currently, Ock, Chan-young leads with the highest number of publications (n = 8), focusing on the application of AI technology in the analysis of tumor-infiltrating lymphocytes (TIL),43 exploring the importance of radiomics analysis in predicting TIL enrichment,45 and the role of the inflammatory immune phenotype (IIP) in the efficacy prediction of ICIs.44 Reck, M, as one of the most frequently cited authors (n = 170), focuses on various treatment methods for non-small cell lung cancer, including immunotherapy, chemotherapy, targeted therapy, and the exploration of biomarkers. These studies have not only enriched our understanding of the treatment of non-small cell lung cancer but also provided valuable insights for personalized medicine. These analysis results emphasize that in the field of ICIs, although research forces are relatively dispersed, individual researchers have already had a significant impact on the knowledge accumulation and clinical application within the field through their high-quality work. The top 10 most cited publications in the field of ICIs combined with AI reveal the research hotspots and priority directions of the field. These papers mostly focus on key issues such as immunotherapy, cancer diagnosis, and machine learning. Co-citation analysis is an effective tool for assessing academic influence, as shown in Table 7, the most cited in this field is the article published in “Nature Biotechnology” in 2019,51 which introduced an innovative machine learning method – CIBERSORTx, an important advancement in digital cell quantification technology. CIBERSORTx is a method that can precisely deduce the gene expression patterns of different cell types from tissue transcriptome data without needing to physically separate the cells. It’s ideal for normalizing variability across platforms and enables broad tissue dissociation analysis using single-cell RNA sequencing. The method’s effectiveness has been confirmed in various tumors, such as melanoma, by examining numerous clinical samples against a single-cell reference map. This analysis uncovers cell-specific phenotypes linked to particular gene mutations and reactions to immune checkpoint blockade. The study’s results not only highlight the potential of digital cell quantification in assessing cellular heterogeneity and deducing cell type-specific gene expression but also showcase its unique advantages, particularly for large sample cohorts and fixed tissue specimens. The co-citation cluster timeline (Figure 7b) further indicates that research in the field of AI applied to ICIs has consistently revolved around clinical applications. As algorithms for image processing technology mature, the research focus has gradually shifted from detailed single factors or single methods (such as TCGA and Pembrolizumab) to a more representative class of diseases (such as pancreatic cancer and NSCLC). Pancreatic cancer, gut microbiome, response assessment, hepatocellular carcinoma, and NSCLC emerged as key research areas around 2019 to 2020, and their popularity has persisted to the present day. The citation burst analysis (Figure 7c) indicates a notable increase in citations for literature from 2018 onwards, signaling the rapid development of AI applications in immune checkpoint inhibitor research over the past seven years. The ongoing high citation rates suggest that combining AI with ICIs will be a burgeoning field, particularly with advancements in machine learning for cancer detection and treatment. Detailed analysis, as depicted in Figure 7a, reveals three main research areas: one on the use of ICIs in NSCLC treatment, especially PD-1/PD-L1 and CTLA-4 pathway drugs; another on bioinformatics and systems biology, including gene expression analysis, cancer genomics, and biomarker discovery for immunotherapy response; and the third on cancer immunotherapy, focusing on the use and efficacy of PD-1 and PD-L1 inhibitors across various cancers. AI is changing cancer treatment, especially for NSCLC, by improving treatment precision through analyzing medical images and biomarkers. Deep learning, a type of AI, can examine CT scans to measure PD-L1 levels, helping to predict how well immunotherapy will work.61 AI can also analyze tissue samples to predict responses to ICIs more accurately than traditional methods.62 AI analysis of gene expression data reveals that non-synonymous mutational burden is linked to the effectiveness of anti-PD-1 therapy, which is crucial for predicting tumor responses to ICIs.63 AI also combines multi-omics data to identify biomarkers related to immunotherapy responses, such as tumor mutational burden, neoantigen load, and cytotoxic marker expression in the immune microenvironment. In cancer immunotherapy, AI is particularly promising for treating head and neck cancer. It accurately measures PD-L1 levels to identify patients who may benefit from ICIs.64 Also, new small molecule inhibitors offer fresh treatment options, potentially leading to better results.65 In summary, using AI with ICIs is a big step forward in cancer treatment. It’s especially helpful for non-small cell lung cancer, bioinformatics, and immunotherapy. AI makes diagnosing easier, finds new biomarkers, and predicts patient outcomes better. In the future, combining AI and bioinformatics might give us more details about how these inhibitors work, helping with early detection and prognosis. Although AI shows great potential in the field of ICIs, it still faces challenges and limitations in practical applications, especially in effectively integrating AI algorithms into clinical treatment plans and prognosis assessments, as well as ensuring data quality and processing standardization. The application of AI in clinical practice is not yet widespread, and although it shows potential in predicting responses to immunotherapy, its accuracy and reliability still need further verification.29 AI models need to handle complex data from different sources and formats, including genomics, proteomics, and radiomics data,66 which increases the difficulty of integrating AI algorithms into clinical decision support systems. To address these issues, future research may focus on several key directions: data collection and preprocessing, feature selection and model training, model validation and optimization, multimodal data integration, model evaluation and clinical application, and continuous monitoring and updating. Creating good AI models needs lots of patient data, such as tumor types, pathology, and biomarker levels. This data is cleaned, standardized, and encoded to train machine learning models like XGBoost, random forests, and SVMs, which analyze the data to find biomarkers and features related to responses to ICIs.67 Combining clinical, proteomics, and gene expression data creates more comprehensive predictive models.68 Integrating AI into clinical decision support systems aids doctors in making personalized treatment recommendations. Regular updates and optimizations of AI models are essential to maintain their predictive performance as new data and research emerge. Ensuring the effectiveness of AI systems hinges on managing the standardization and quality of data. Medical data is complex and diverse, and noise issues make the task challenging.69 For instance, in digital pathology, AI has to deal with tissue slide images from various sources, which affects data consistency and interoperability.29 At the same time, AI models need to be explainable so that doctors can trust and utilize AI’s predictive outcomes.70 To enhance the accuracy and interoperability of medical data, several techniques and methods have been developed. First, standardize the terminology and frameworks for data quality assessment to understand whether electronic health records are suitable for specific purposes.71 Second, we establish frameworks for managing the quality of health data, including quality checks and assessments of complex health data.72 I In the era of big data, standardizing lab data is also crucial, which helps ensure data compatibility and reliability across different systems.73 The fusion of AI with ICIs is set to usher in an era of personalized and precise medicine. Despite existing challenges, the hope it brings for improving the effectiveness of cancer treatment and the quality of life for patients is evident. Limitation In our study, we recognized several significant limitations. First, our literature search was primarily based on the SCI-E and SSCI indexed categories within the Web of Science Core Collection. Although we carefully selected the indices most closely related to our research topic, this approach might have missed relevant literature from other databases, thereby introducing selective bias. Moreover, due to specific restrictions of the databases, certain countries were disproportionately represented in our research findings, inevitably leading to a certain degree of bias. At the same time, constrained by the algorithms of the bibliometric software, we had to perform a certain degree of data filtering and aggregation during parameter setting and data processing, which might introduce some bias. Despite our best efforts to minimize this impact, we cannot entirely rule out the possibility. Second, our study was limited to English-language literature, which might have overlooked valuable findings in non-English literature, leading to incomplete data. Additionally, our bibliometric approach relied on publication counts and citation frequencies, metrics that, as indicators of research field impact, may inherently be biased. Lastly, since the accumulation of citation counts is a process that unfolds over time, our study may not have fully captured the impact of recently published literature, which could also affect the accuracy of our assessment results. Despite these limitations, our study strives to comprehensively cover the existing publications on the application of AI in the field of ICIs and provide valuable insights into the research hotspots, challenges, and future development directions in this field. Conclusion This study used bibliometric methods to analyze AI’s role in ICIs research, showing AI’s growing importance in treating non-small cell lung cancer, bioinformatics, systems biology, and cancer immunotherapy. The focus is shifting toward precise diagnostics and disease predictions. China leads in publications, but the US has a bigger impact. Strengthening international cooperation, especially with scientific leaders like the US, is key to advancing the field. Interdisciplinary collaboration is also important. Combining expertise from computer science, bioinformatics, genomics, and clinical medicine can lead to new ideas and more effective AI models for ICIs. This can improve diagnostic and treatment accuracy and prognosis, advancing science and enhancing patient care.