On July 14, 2023, we retrieved all citation data published between January 1, 1900, and July 14, 2023, from the Web of Science Core Collection (WoSCC). We selected WoSCC because it serves as a comprehensive and standardized collection of databases widely utilized in academic circles (22). It encompasses a broader range of research fields than PubMed and spans research from 1900 to the present. This analysis focuses on articles published from 1900 to 2023 in peer-reviewed journals. The detailed search string can be found in Figure 1 . The document type included articles and reviews. We gathered essential data for each publication, including the title, abstract, authors, institution, country or region, journal, keywords, and references.

Unpublished documents and document types other than articles were excluded. The citation data were downloaded on July 14, 2023, and some 2022 documents included by WoSCC were not published. Thus, they were excluded from this study. Data with document types other than articles, such as procedures, papers, review articles, meeting abstracts, early access, editorial materials, book chapters, letters, corrections, data papers, books, and retracted publications, were also excluded. Figure 1 presents the detailed search, exclusion, and analysis processes.

This study employed bibliometric analysis to explore global research trends in the application of AI in skin cancer. Collaborative networks among countries, institutions, journals, keywords, and research categories were analyzed and visualized using CiteSpace (version 6.2. R3), VOSviewer (version 1.6.18), SCImago, Microsoft Excel 2019, and R 4.2.3. The study collected citation data, encompassing the yearly publication count, countries, institutes, journals, and keywords.

The literature retrieval revealed that research on AI in this topic commenced in 1991. Between 1991 and 2023, 512 papers have been published, allowing us to identify publication trends related to AI in skin cancer research (see Figure 2 ). Studies in this area have been steadily increasing, signifying the establishment of a significant research trend. The annual number of publications in this field has been on the rise, particularly after 2016, experiencing a rapid surge, reaching 159 publications in 2020. This suggests that this research field has garnered increasing attention from researchers in recent years, emerging as a focal point.

A total of 2,539 authors and 13,055 co-cited authors were included in the study. Table 1 lists the top 10 most productive authors, with Brinker, Titus J, Hekler Achim, and Schadendorf Dirk ranking as the top 3 authors, with 20, 17, and 16 articles, respectively.

According to Price’s law, the minimum number of papers that core authors should publish in a specific field is represented by N, where N = 0.749 × ηmax 1/2. Here, ηmax is the number of publications by the most prolific author. VOSviewer statistics indicate that ηmax = 20, leading to an approximate value of 3. Hence, authors who have published over three papers are considered core authors. A total of 73 core authors have collectively contributed to 307 papers, meeting the 50% standard proposed by Price. Consequently, it can be inferred that a relatively stable cooperative group of authors has coalesced within this field. CiteSpace was employed to visualize the co-authorship map of authors (see Figure 2 ), and the co-occurrence map of authors’ cooperation network is presented in Figure 3A .

We identified 70 countries or regions with published studies on skin cancer; an overview of the collaboration among these entities is presented in Figure 3B , and the top 10 countries are detailed in Table 2 . The United States leads in the number of studies published (149), followed by India (61), Germany (57), and China (48). Figure 3B highlights the intricate and robust collaborative relationships among different countries. We employed the CiteSpace platform to assess the centrality of countries. As indicated by the color gradient, the bold yellow line in Figure 3B represents the most recent partnerships between countries. The size of each circle corresponds to its centrality, with larger circles indicating higher centrality values. The various colored lines connecting the nodes represent collaborations in different years between countries or institutions. The United States (centrality = 0.33), Germany (centrality = 0.25), and China (centrality = 0.21) exhibit extensive collaboration with other countries.

We identified 1,127 institutions contributing to this field, with the top 10 listed in Table 3 . Ruprecht Karls University Heidelberg (26) leads, followed by the Helmholtz Association (25), the German Cancer Research Center (DKFZ) (24), and the Memorial Sloan Kettering Cancer Center (22). Data analysis revealed that eight of the top 10 institutions are located in Germany, while the remaining two are in the United States. Figure 4 illustrates the cooperative relationships among institutions, indicating that the collaboration is both widespread and partly concentrated.

It is evident that there is a need to strengthen both intra-regional and international exchanges and cooperation. The findings indicate that medical colleges and hospitals constituted a relatively significant proportion within the research domain, serving as the primary drivers of research and publication. Inter-institutional collaboration is predominantly confined to local regions, with limited cross-regional cooperation observed. These results underscore the importance of future teams avoiding singular institutional or regional focus and, instead, emphasizing enhanced inter-team collaboration.

In any research field, the referential relationships among academic journals typically reflect knowledge exchange, citing studies representing knowledge frontiers and referenced studies forming the knowledge foundation. Table 4 displays the top 10 citing journals, with the most frequently cited journals in the included references being Cancer (21 citations), Diagnostics (19 citations), and the European Journal of Cancer (17 citations).

Dual-map overlay visualization can illustrate the distribution of papers in each subject, citation trajectory, the center of gravity shifts, and other relevant information (23). Figure 5 presents a dual-map overlay of journals, showcasing citing and cited journals on the left and right, respectively, with citation relationships represented by colored paths.

There were primarily four citation paths, with citing papers mainly concentrated in four fields: 1) molecular biology and immunology; 2) medicine, medical, and clinical disciplines; 3) mathematics, systems, and mathematical studies; and 4) dentistry, dermatology, and surgery. The cited papers were mainly situated in four fields: 1) systems, computing, and computer science; 2) molecular biology and genetics; 3) health, nursing, and medicine; and 4) dermatology, dentistry, and surgery.

In a research field, a hot topic is characterized by a high frequency of keywords. In contrast, high-centrality keywords indicate the location and significance of associated research content within that field. To conduct a keyword analysis of global research trends in the application of AI in skin cancer, we retrieved and examined relevant literature to identify the most frequently used keywords.

The top 10 keywords are presented in Table 5 , which can be categorized into three topics for further discussion. i) Research subject: “skin cancer” and “melanoma” have the highest frequency of occurrence. This is attributed to melanoma having the highest fatality rate and is one of the most aggressive and lethal forms of cancer, with increasing worldwide incidence in recent decades (24). ii) Research purpose: the keyword “diagnosis”, which first appeared in 1992, reflects the primary objective of applying AI to skin cancer. Meanwhile, the emergence of “classification” and “segmentation” since 2003 indicates more refined objectives in the diagnostic process. The high frequency of these keywords is closely related to developing relevant image-processing algorithms. iii) Research techniques: “machine learning” first appeared in 2003. Machine learning is a significant branch of AI, and strictly speaking, both “deep learning” and “CNN” fall under the umbrella of machine learning algorithms (25). Interestingly, we observed that “neural networks” appeared in related literature as early as 1994, while “deep learning” and “CNN”, derived from this concept, only began to frequently appear decades later, in 2016 and 2019, respectively. This aligns with the trajectory of AI development, where recent innovations in computer hardware technology have revitalized past theories. This insight suggests that future breakthroughs in computer hardware technology may play a critical role in advancing AI for clinical applications in skin cancer diagnosis. However, future research on more generalized AI algorithms applicable to various types of skin cancer diagnosis may hold even greater value.

Keyword clusters can illustrate the structural system of related research fields. The results of keywords clustering revealed Q = 0.6142 > 0.3 and S = 0.8309 > 0.7 in Figure 6 , indicating the significance of the cluster structure and the credibility of the results (26). The timeline map visualizes the number of keywords within each cluster. A cluster’s significance is determined by the number of keywords it contains, and it also reveals the temporal span of keywords within each cluster. This facilitates the examination of the emergence, growth, and decline of specific research clusters, thereby assisting in exploring temporal patterns characterizing the research field represented by these clusters (27). We obtained the timeline view for the 13 clusters in Figure 7 , where keywords from the same cluster are placed on the same horizontal line. The chronological occurrence of keywords is positioned at the top of the view, with the timing progressing toward the right. “#2 Artificial intelligence” first appeared in 1992 and was the earliest keyword. The research focused on “#7 Radiomics”, which appeared latest among the 13 clusters. The timeline for “#0 Dermatologists”, “#1 Skin lesions”, “#2 Artificial intelligence”, “#4 Optical coherence tomography”, “#7 Ensemble learning”, “#11 Radiomics”, and “#12 Neural network” indicates that they are closest to 2023. This suggests that these subjects have recently gained more attention and are likely to become research hotspots soon.

The analysis of emerging burst keywords can reflect the changing trends in the field’s hot topics, further supporting the findings of this study. The strength value indicates the frequency of citations. The blue line represents the time interval, while the red line represents the duration of the citation burst. In Figure 8 , it is evident that “Pigmented skin lesions”, “Image analysis”, and “Epiluminescence microscopy” exhibited the highest outburst intensity among keywords. Additionally, we identified the emergence of new keywords, depicted by the red grid, with “Feature extraction”, “Algorithms”, “diagnostic accuracy”, “level classification”, superior”, and “texture” being particularly popular from 2019 to 2021. Notably, “Image analysis” had the most extended usage from 2003 to 2019. Our keyword analysis offers valuable insights into the most frequently used and emerging keywords in AI application research for skin cancer.

In this study, 512 papers, meeting the search terms and adhering to the inclusion/exclusion criteria, were published between 1991 and 2023. A corresponding network diagram was constructed for visual analysis, aiming to understand the dynamic evolution process of core researchers, research focal points, and research frontiers in applying AI to skin disease research. This analysis also serves to explore the developmental trends within this field.

i.The consistent year-on-year increase in publications within this field, notably surging after 2016, aligns with the recognition of the open AI ecosystem as one of the 10 most pivotal emerging technologies by the World Economic Forum in the same year. This suggests a significant correlation between these trends (31). There is indeed a correlation between them. The top three authors were Brinker Titus J (20), Hekler Achim (17), and Schadendorf Dirk (16). The United States published the highest number of studies (149), followed by India (61). Germany claims eight positions among the top 10 institutions, while the United States holds two. The most frequently cited journals in the included references were Cancer (21 citations), Diagnostics (19 citations), and European Journal of Cancer (17 citations).

ii.Intra-regional and international exchanges and collaboration must be bolstered. The findings unveiled that medical colleges and hospitals comprised a significant portion of the research domain, acting as the principal driving forces behind research and publication. Inter-institutional collaboration was predominantly confined to local regions, with minimal cross-regional association observed. These results indicate that future teams should endeavor to steer clear of singular institutional or regional focus instead of prioritizing enhanced inter-team collaboration.

The incidence of skin cancer is progressively rising each year. Despite its comparatively low mortality rate, it continues to place a substantial financial burden on health services. It can lead to severe mental health issues, mainly because most skin cancers manifest in highly conspicuous areas of the body. Due to limited awareness of screening, the absence of distinct lesion characteristics in early-stage skin cancer, and inadequate clinical proficiency and services, most patients receive diagnoses at advanced stages, leading to unfavorable prognoses. Hence, there exists an immediate necessity for AI systems to aid healthcare professionals in this field (32).

AI is a branch of computer science that enables machines to perform human-like tasks through learning from experience (33). Machine learning, a subfield of AI, focuses on training computers to enhance their performance in specific tasks through experiential learning (26). In machine learning, algorithms self-train by directly analyzing data to discern patterns rather than composing software with explicit instructions for a particular task. Two common types of machine learning tasks are supervised and unsupervised learning (34, 35). Supervised learning involves algorithms utilizing labeled training, entails data classification, and programming the relationship between input and output data. Algorithms for classification in supervised learning include ANNs, decision tree networks, support vector machines, and Bayesian networks. Unsupervised learning employs algorithms to discern hidden patterns within datasets, yielding diverse outcomes such as cluster analysis, dimension reduction, and association. Deep learning, a subset of machine learning, is rooted in deep neural network architectures. Natural language processing is another machine learning application that employs software programming to comprehend and manipulate natural language text or speech for practical purposes (33). Currently, AI finds extensive utility in skin cancer across various aspects.

Research into the application of AI in the diagnosis and management of skin cancer has witnessed rapid growth in recent years, unveiling numerous promising avenues for future investigations. Much of this research has centered on melanoma, a skin cancer with a high mortality rate that has garnered extensive attention. Early screening and diagnosis are pivotal in enhancing patient survival, especially given the time-consuming nature of manual examinations and the potential for confusion with benign skin lesions, even among seasoned dermatologists. Moreover, the disease diagnosis can be particularly challenging in regions with limited medical resources. To address these challenges, expanding skin disease image databases to encompass a more comprehensive array of skin cancer types and other complex-to-diagnose benign conditions may offer valuable clinical support in the future.

Although AI’s role in analyzing skin mirror images and skin pathological images has been extensively discussed, fewer reports on AI research leverage other types of photographic samples. Researchers can delve into additional sample types to tackle factors such as variations in zoom, angle, and lighting conditions, which can significantly impact the accuracy of clinical image analysis.

Moreover, the technological advancements in AI that have been harnessed in the context of other diseases are increasingly finding application in skin cancer. This includes the implementation of personalized and precision medicine approaches that draw from diverse data sources, encompassing genetic databases, medical records, tissue banks, and various clinical databases. The influx of data, often called “big data”, necessitates robust computing power for efficient processing and analysis. AI, coupled with the exponential growth in computing power capabilities, augments our analytical capacity and facilitates the seamless integration of various datasets to extract meaningful insights. Precision medicine relies on extensive population-level data to inform individualized treatment decisions, and AI plays a pivotal role in navigating and interpreting this intricate landscape (48, 49).

Much of the existing research compares dermatologists with AI, but these two entities do not have a hostile relationship. AI does not aim to replace human doctors in making independent diagnoses; instead, it is a valuable tool for assisting doctors in diagnosis and treatment. One study, for instance, investigated the potential benefits of integrating human expertise with artificial intelligence for the classification of skin cancer (50). Dermatologists should not perceive AI as a threat to their professional knowledge; instead, they should consider it a complementary resource in clinical practice for the foreseeable future. Therefore, future research should continue to explore the synergy between humans and computers in the field. By better understanding AI concepts, dermatologists can enhance their ability to provide improved patient care and treatment.

AI faces several challenges in medical applications: 1) complex clinical scenarios characterized by inconsistent disease classification, a lack of standardized diagnostic criteria, and subjective evaluation standards can pose significant hurdles when designing disease diagnosis systems. This complexity may result in variations in the performance of intelligent algorithms (51). 2) The relatively limited sample size available for training AI models, coupled with incomplete and fragmented databases, along with variations among tested individuals, raises legitimate concerns regarding the actual effectiveness of AI systems. 3) Current auxiliary diagnostic systems still lack the capability for independent diagnosis, while the interpretability of intelligent diagnostic systems falls short of the desired standards. Therefore, further validation and confirmation of their efficacy become imperative (52).

To address these challenges and promote the advancement of AI in diagnosis, the following recommendations are proposed: 1) establishing standardized disease classifications and diagnostic criteria: to enhance the development of more robust and practical diagnostic systems, it is imperative to establish standardized disease classifications and diagnostic criteria. This will reduce the reliance on subjective standards, ensuring the generality and reliability of these systems. Active involvement of experienced clinical professionals in the development process is paramount. 2) Curative diverse and extensive datasets: accessing diverse and extensive datasets is essential to refine AI models effectively. Creating a unified and high-quality image database that can be continuously optimized and expanded to meet the diagnostic demands of various diseases is vital. This approach helps overcome the limitations associated with sample size. 3) Strengthening professional training: maximizing the utilization rate of AI technology necessitates strengthening training programs for relevant professionals. Ensuring that healthcare professionals are well-equipped to integrate AI into their practice is crucial for successfully adopting and implementing AI-based diagnostic tools. These recommendations aim to propel the field of AI in diagnosis forward, fostering innovation and improved healthcare outcomes while adhering to rigorous scientific and professional standards.

The application of AI in skin cancer diagnosis and management represents a dynamically evolving field with numerous promising research avenues. By persistently exploring these directions, researchers can forge innovative solutions aimed at enhancing the precision, efficiency, and availability of skin cancer diagnosis, all while conscientiously addressing the ethical and legal dimensions of this undertaking. To establish the foundation for the legal recognition of AI-based diagnosis, researchers should undertake concerted efforts to advance the technology and secure the endorsements for autonomous diagnosis when the technology reaches an adequate level of maturity.