Lung cancer stands as the leading cause of cancer-related deaths globally and ranks as the second most commonly diagnosed cancer. The average 5-year survival rate is only 15% (1).With concerted efforts to advance CT early screening and update treatment methods, the latest epidemiological report on lung cancer in the United States indicates a continuous decline in the incidence rate and mortality of lung cancer (2).Consequently, the development of earlier and more accurate diagnosis, along with more precise and personalized treatment, holds significant importance for the prevention and prognosis of lung cancer.

Artificial Intelligence is a discipline that mainly studies the application of computers to simulate human intelligent behavior, involving various disciplines such as computation, mathematics, biology, etc (3).With the arrival of the big data era and the updates and progress of computer equipment and algorithm technology, artificial intelligence has been widely applied in multiple fields, gradually penetrating into all aspects of our lives. In the 1950s, the medical field had already noticed the potential of AI and began to attempt to apply AI to auxiliary diagnosis of diseases (4). In recent years, AI has gradually been widely applied in various aspects of healthcare with its powerful algorithms and learning capabilities, including disease diagnosis, prognosis prediction, drug research, genomics data analysis, etc., bringing new methods such as imaging omics, pathomics, genomics, etc. to the medical field (5). A mount of AI technologies such as machine learning (ML) and deep learning (DL) have been used for auxiliary diagnosis and prognosis prediction of lung cancer and achieved good predictive performance (6–9).More and more scholars are beginning to explore the application and implementation of AI in the field of lung cancer, and the number of research studies in this area is also growing exponentially. This makes it increasingly difficult for most researchers to keep up with the latest research findings, stay informed about research trends, and anticipate future developments.

Bibliometric analysis is an information visualization method that involves summarizing all literature globally in a specific field. It utilizes mathematical and statistical methods to quantitatively analyze bibliographic data and measurement characteristics. This process aims to comprehend the knowledge structure of a particular field and identify research frontiers or hotspots, representing a form of information visualization. Due to the rigorous and objective nature of bibliometric analysis, scholars in various fields commonly employ this method to conduct research in their respective domains (10). However, as of now, there are no bibliometric studies related to AI in lung cancer. Therefore, we aim to conduct a quantitative and qualitative analysis, along with visualization, of the research progress and current status in the field of artificial intelligence applied to lung cancer by collecting relevant literature from databases over the past 20 years. This endeavor is intended to provide insights into potential future research trends, aiding scholars in this field to develop a more systematic understanding of research priorities and future directions.

With the advancement of hard drives and semiconductors, the capabilities of big data storage and dataset-based computer modeling have become more potent, paving the way for the goal of computers that can simulate human interaction (4). In tandem, AI technology has emerged and rapidly evolved, finding applications across various disciplines. The field of lung cancer, being one of the world’s challenges, is no exception. Bibliometrics allows for the analysis of authors, institutions, countries, and references in the WOSCC bibliographic database, providing insights into a specific research area and visualizing it through tools like Citespace and VOSviewer. This research methodology offers a more comprehensive analysis of literature and presents more visual results than a typical systematic review. In the realm of AI in lung cancer, this study employs bibliometrics as its initial approach to explore the applications and developments in the field over the past two decades and to speculate on future research trends.

In the initial stages, the development of AI in lung cancer research was slow, with an annual publication output of fewer than 30 papers before 2015. After an exploratory period from 2015 to 2017, there was a steady increase, and the growth rate exceeded 100 papers per year. It is projected that by 2023, the annual publication output in this field will surpass 1000 papers ( Figure 2 ). This phenomenon indicates that the field is currently experiencing rapid development. The year 2017 marks a crucial turning point in the development of AI in LC, representing a culmination of earlier accumulations and aligning with the trends of the era. On one hand, the emergence of DL and convolutional neural networks (CNN) has led to breakthroughs in processing techniques, especially for complex data such as medical images (26). Simultaneously, the establishment of large-scale public databases related to LC, such as The Cancer Genome Atlas (TCGA), The Cancer Imaging Archive (TCIA), and the LIDC/IDRI (11) Lung Nodule CT Image Database, has facilitated the prominence of genomics and radiomics. On the other hand, in 2017, the victory of AlphaGo over world Go champion Lee Sedol garnered global attention and admiration. Major economic powers worldwide accelerated their deployment of artificial intelligence, propelling the climax of AI development to new heights (27). The attention to AI research in the LC field also increased, with an annual average growth rate of 70.39%, particularly evident in 2020 ( Figure 2 ). Although the publication output for 2023 is not fully accounted for due to the study’s timeline, based on the current trends, the application of AI in LC is expected to remain a future research hotspot and will continue to receive attention.

The top two countries in terms of publication output in this field are China and the United States, aligning with the national rankings for lung cancer incidence (1). These two countries contribute to over 60% of the total global publication output, revealing significant research disparities in this field among countries worldwide. China and the United States hold a decisive advantage compared to other nations. Chinese scholars have a total publication output of 1173 papers, but the average citation frequency per paper is relatively low at 15.7 times per paper, similar to other Asian countries such as South Korea and India. However, there is still a gap compared to European and American countries, indicating that the quality of Chinese papers needs improvement compared to their Western counterparts. Through in-depth analysis of the data, it was found that China’s annual publication output began to grow rapidly only in 2017 and surpassed the United States after 2019, reaching twice that of the United States by 2022. This suggests that China started later in this field but has developed rapidly, which may be one of the reasons for the lower average citation frequency. The United States ranks second in publication output but first in total citations, indicating its central position in this field. However, publication output and citation rates are just indicators of research impact, and careful interpretation should consider other factors such as the primary research language, international collaboration, academic achievement policies, language bias in indicators, and publication bias of journals.

Figures 3C, D illustrate cooperation between different countries. Among the top 10 countries in this field, most research from countries other than the Netherlands is conducted as single-country studies, indicating a need for stronger international collaboration in AI in lung cancer. There is extensive collaboration between China and the United States, radiating to the North America, East Asia, and Australia regions, indicating that China and the United States, as leading countries in this field, have significant international appeal and influence. Close collaboration is observed among European countries like the Netherlands, Germany, England, and Italy, while strong collaboration exists between Asian countries like India and South Korea. This suggests that collaboration tends to be regional, but there is less collaboration between regions, likely influenced by language and geographical factors.

Among the top 10 institutions, China has 7, while the United States, the Netherlands, and South Korea each have 1. China boasts numerous research institutions and publications, largely due to the robust support for artificial intelligence applications from the Chinese government in recent years. This support spans across various fields involving artificial intelligence (28–31), indicating that China’s research in this field may lead in the future with continuous investment. However, it’s noteworthy that Maastricht University and Harvard Medical School have significantly higher average citation counts of 68.07 and 66.94, respectively, far surpassing other institutions in the top ten. This suggests that they are central institutions in this field. This distinction is mainly attributed to the contributions of Professor Philippe Lambin’s team at Maastricht University and Professor Hugo J.W.L.Aerts’s team at Harvard Medical School. They jointly published the groundbreaking paper “Radiomics: extracting more information from medical images using advanced feature analysis” in 2012, introducing the concept of “radiomics” and ushering in a new era for the application of artificial intelligence in medical imaging (12). Subsequently, they successively published a series of studies (32–35) applying machine learning and deep learning to radiomics in lung cancer, which garnered widespread sharing and citation.

Generally, authors with higher citation counts are considered to have a greater impact than those with lower citation counts, and authors who are co-cited may be concentrated in related research areas. Looking at author contributions and co-cited authors ( Table 4 ), the most prolific author in this field is Qian W, with 20 publications, primarily focused on Cancer Computer-Aided Diagnosis (CAD) systems. When considering co-cited authors, Armato SG is the most frequently cited and central co-cited author, indicating his significant role in this field. It’s noteworthy that more than half of the active top 10 authors come from the Asian region (4 from China, 2 from South Korea), but all of the top 10 co-cited authors are from Western countries. Furthermore, in the comprehensive analysis of institution and author collaboration networks ( Figures 4C and 5C ), we observe that collaboration among authors in the Asian region is mostly limited to the same country. Even on a global scale, only a few highly influential institutions and authors engage in international collaboration. This suggests that research in the field of AI in lung cancer is still relatively dispersed, lacking international exchange. International collaboration and author influence are complementary, and authors such as Philippe Lambin, Hugo J.W.L.Aerts, Issam EL Naqa, and Weiming Li are successful figures in the field, significantly influencing other authors. Their teams could be excellent potential collaborators for researchers.

Publications are the carriers of research achievements, and effective scientific communication requires publishing research results in internationally peer-reviewed journals. Therefore, through the analysis of the distribution of journal sources, researchers can quickly identify the journals most suitable for their papers (36). The analysis of journals and co-cited journals ( Table 7 ) reveals that, except for the International Journal of Imaging Systems and Technology, the top 20 journals are all excellent journals with a JAR ranking of Q2 and above. Among them, the top 3 journals in terms of publication volume are Frontiers in Oncology, Scientific Reports and Cancers all with over 80 publications, significantly higher than other journals. This indicates that these journals prioritize research in this field, and scholars in the field can give priority to publishing their findings in these journals. In the field of Radiology and Imaging, the top journal is Radiology with the highest co-citation frequency. Additionally, other highly influential medical journals such as New England Journal of Medicine, Nature and CA-A Cancer Journal for Clinicians are also listed. This, to some extent, reflects that the application of artificial intelligence is an important research direction in the field of lung cancer. Regarding the distribution across disciplines, in addition to medical-related journals, there are also journals in the fields of computer science and engineering. This indicates that the application of AI in LC is an interdisciplinary field that requires collaboration across multiple disciplines.

The analysis of keywords provides another perspective on the development process and trends in the field. To gain a macroscopic understanding of the research hotspots and frontiers of AI in LC, we conducted a visual analysis of high-frequency keywords ( Figure 6 ). Representative keywords include “lung cancer,” “classification,” “cancer,” “machine learning,” “deep learning,” and “diagnosis,” indicating that these topics are the research hotspots in this field. Currently, AI applications in LC mainly focus on identification, diagnosis, and therapeutic prediction, with machine learning and deep learning being the most commonly used methods.

Further keyword clustering analysis using CiteSpace was conducted. Noun terms were extracted from the titles of cited literature as labels for clustering, employing the Log-Likelihood Ratio (LLR) algorithm as the extraction method. As shown in Figure 7A , the keyword cluster view indicates that “cell lung cancer” #0 and “deep learning” #1 are the largest clusters, suggesting that the application of deep learning in non-small cell lung cancer may be a mature and significant topic in this research field. Currently, deep learning has been widely applied in the clinical diagnosis (16, 37, 38), treatment (35, 39) and prognosis prediction (34, 40) of lung cancer. Simultaneously, the Timeline View analysis ( Figure 7B ) reveals that AI in the field of lung cancer has consistently focused on clinical applications. As detection devices advance and treatment methods evolve, research priorities gradually shift towards early screening, immunotherapy, risk prediction, and other areas. After undergoing early theoretical research and technological exploration, AI has been widely applied in various fields related to lung cancer, including CT imaging, pathological images, genomics, etc., achieving encouraging results and providing accurate guidance and support for clinical early diagnosis and treatment decisions. The duration of keyword prominence was longer before 2016, but became shorter after 2016 ( Figure 7C ). This phenomenon indicates a slow development of AI in LC before 2016, followed by a rapid development phase after 2016, attributed to accelerated technological iterations leading to shorter bursts of prominence. Additionally, emerging terms post-2020, such as shape, volume, radio genomics, predominantly focus on the analysis of imaging data. This suggests that CT and PET/CT, as the most crucial non-invasive diagnostic tools for lung cancer, still hold significant research value. Imaging genomics and related multi-omics studies are identified as the forefront direction in AI research for lung cancer.

Overall, the application of AI in LC can be broadly categorized into three directions. The first category involves the application in early screening and identification of lung cancer. Detection of pulmonary nodules is crucial in low-dose CT screening for lung cancer, and efficient detection significantly enhances the risk assessment of lung cancer. Jiang et al. (41) designed a four-channel convolutional neural network model based on multiple sets of patches cut from lung CT images, which effectively detects pulmonary nodules. Compared to solid solitary nodules, ground-glass opacity (GGO) nodules are more likely to be malignant. He et al. (42) used a 3D CNN to detect the position of GGO nodules and classify lesions (benign or malignant), achieving a competition performance metric (CPM) of 0.817. A portion of metabolic by-products released by human tissues enters the blood, undergoes substance exchange in the lungs, and is expelled from the body through the respiratory tract. Therefore, exhaled breath to some extent can reflect the body’s disease status. The relationship between volatile organic compounds (VOCs) in exhaled breath and lung cancer is a focus of research (43, 44). Researchers defined an instrument, known as an electronic nose (45), composed of an array of electronically sensitive sensors and a pattern recognition system capable of identifying simple or complex odors. In the study (46), the electronic nose was used to distinguish between lung cancer patients and healthy individuals, achieving a discrimination sensitivity of 81% and specificity of 91%. The highest sensitivity, reaching 92%, was observed in stage I lung cancer.

The second category of application involves the use of AI in lung cancer classification. Son et al. (47) found that radiomic features contribute to differentiating invasive adenocarcinoma from in situ and minimally invasive adenocarcinoma. Wu et al. (33) employed machine learning methods to explore the predictive performance of radiomic features for lung cancer histological subtyping (adenocarcinoma and squamous cell carcinoma). The results showed that 53 radiomic features were significantly correlated with lung cancer histological subtypes, indicating substantial potential for radiomic features in predicting lung cancer histological subtypes.

The third category involves the application of AI in the prognosis of lung cancer treatment. Deng et al. (48) developed a model based on deep learning and pre-treatment CT for a multicenter prognostic study to predict the survival benefits of epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKI) and immune checkpoint inhibitors (ICI) in stage IV non-small cell lung cancer (NSCLC) patients. The model increased the diagnostic accuracy of clinicians with two years of experience from 47.91% to 66.32% and clinicians with five years of experience from 53.12% to 61.41%. Wang et al. (49) developed a prognosis model based on DL and pathological images of lung adenocarcinoma, achieving favorable predictive performance. Cui et al. (50) integrated multi-omics information into the actuarial deep learning neural network (ADNN) architecture for the joint prediction of radiotherapy outcomes, radiation pneumonitis, and local control in stage III NSCLC patients. The results surpassed traditional normal tissue complication probability/tumor control probability models (C-index = 0.660 vs 0.613/0.569).

These studies demonstrate that machine learning and deep learning are currently the most commonly used methods. With the optimization of algorithms and the progress of multi-center studies involving multiple omics, the predictive capabilities of AI have significantly improved. This enhancement effectively increases the clinical diagnostic efficiency and prognosis accuracy of lung cancer, assisting physicians in making correct clinical decisions.

Based on the above analysis, our bibliometric study systematically analyzed the basic situation, research hotspots, and trends of Artificial Intelligence in the field of Lung Cancer from a visual perspective. Therefore, the results of this bibliometric study are objective and accurate, providing comprehensive guidance for clinical physicians and scholars engaged in research in this field. Given the global expansion of digital networks and the continuous innovation of AI technology, the role and advantages of AI in clinical diagnosis and prognosis prediction for LC are becoming increasingly prominent. The application of AI in the field of LC is undoubtedly a current research hotspot and a major research direction for scholars in the coming years.

However, the application of artificial intelligence technology in the field of lung cancer still faces some limitations and challenges. AI requires large sample sizes to support its applications, with the demand for natural images in public domains often reaching tens or even hundreds of thousands. For instance, the Objectron, an open-source 3D object dataset released by Google AI in 2020, already contains 15,000 short video samples and over 4 million annotated images collected from five continents and ten countries. In contrast, in the medical field where higher precision is required, AI datasets typically consist of fewer than a thousand cases and are often retrospective. While data augmentation methods such as flipping and cropping can be employed to expand the dataset (51), prospective multicenter studies with large samples are crucial to demonstrate the reliability of AI models and their clinical utility in the real world.

The protection of patient privacy information limits data collection and sharing. Methods like the Three-Dimensional Shearlet Intuition Fuzzy Algorithm (STIF) (52) have been introduced to address this issue. Additionally, due to variations in data sources, including differences in race, disease severity, and acquisition device parameters, algorithms may vary significantly (53). Therefore, international consensus is needed to guide methodology, and international multicenter studies with large samples are necessary to validate the accuracy of models.

The end-to-end nature of deep learning obscures the data processing process within the model, leading to a lack of interpretability. Before deploying deep learning-based lung cancer models in clinical practice, legal responsibilities and patient safety issues must be considered. Moreover, the complexity of deep learning algorithms implies high hardware requirements and a need for expertise in computer science. Therefore, fostering communication between computer professionals and medical experts and promoting high-quality collaboration between the field are essential.

There are several noteworthy limitations to our study. Firstly, our reliance solely on the WoSCC database implies the potential omission of relevant papers from other databases (21). However, integrating various databases for analysis poses challenges due to limitations in bibliometric software. We have elucidated our rationale for selecting WoSCC as our primary database in the Methods section. Secondly, the implementation of stringent search strategies and the restriction to English-language literature may result in data loss, introducing research bias and diminishing credibility. Lastly, the dynamic nature of databases, coupled with the timing of our study, raises the possibility of underestimating the impact of recently published high-quality articles due to suboptimal citations (54, 55). Additionally, some literature from 2023 may not have been incorporated into our study. Despite these limitations, our study encompasses the majority of publications on AI applied to LC, providing valuable insights into current research hotspots, evolutionary processes, and trends in this field.

In conclusion, the widespread application of artificial intelligence in the realm of lung cancer is particularly pronounced in auxiliary diagnosis and prognosis prediction. Research emphasis in this domain is progressively shifting towards non-invasive diagnosis and precision therapy facilitated by deep learning techniques. Notably, China and the United States stand as frontrunners in this field, likely maintaining their leadership positions for the foreseeable future. Nevertheless, there is a need for heightened transnational collaboration, particularly among Asian countries, which should actively cultivate close partnerships with developed nations such as the United States and the Netherlands. Simultaneously, there is a call for enhanced interdisciplinary collaboration between the fields of medicine and computer engineering.

The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.

RZ: Data curation, Formal Analysis, Writing – original draft, Writing – review & editing. TG: Software, Visualization, Writing – original draft, Writing – review & editing. JL: Data curation, Formal Analysis, Writing – original draft. ZL: Software, Writing – original draft. XT: Writing – review & editing, Funding acquisition, Validation. CZ: Methodology, Visualization, Writing – review & editing. XL: Data curation, Writing – review & editing. YW: Writing – review & editing. LG: Funding acquisition, Writing – review & editing. KH: Funding acquisition, Writing – review & editing.