Global trends in research on Artificial Intelligence use in cariology: a bibliometric and altimetric review

Introduction

Artificial intelligence (AI) and its applications, particularly in the field of reinforcement learning, are becoming increasingly important in dentistry, as highlighted by Chen et al.¹ This technology aims to assist, support and, in some cases, replace traditional human activities.² Key subsets of AI include machine learning, neural networks and deep learning.³ These technologies enable machines to develop algorithms that perform tasks and solve problems autonomously, without human intervention.⁴ Growing interest in clinical applications of AI has led to research focusing on its use in cariology.⁵

Despite the widespread availability of preventive measures, untreated dental caries affected 2.3 billion people worldwide in 2017, making it the most prevalent health condition globally.⁶ Initial caries lesions present as white spots that can progress to dentin cavitations, potentially affecting pulp vitality and leading to tooth structure loss in severe cases.⁷ The diverse presentation and location of caries within the dental structure pose significant challenges for early diagnosis.⁸ This diagnostic complexity increases the risk of tooth loss and elevates healthcare costs due to the need for clinical materials and exposure to radiation from frequent radiographic imaging.⁹

In response to these challenges, artificial intelligence (AI) has emerged as a promising technology to enhance diagnostic accuracy, reduce overdiagnosis, and minimise radiation exposure.^1,9 By autonomously processing large volumes of data, AI can facilitate earlier and more precise detection of caries, potentially decreasing the need for invasive procedures and lowering associated healthcare costs.⁴ Consequently, the integration of AI into dental practice has the potential to benefit both patients and healthcare systems by providing more effective and accessible diagnostic tools, though further studies are needed to fully understand its long-term impact and efficacy.⁹

Bibliometric analysis can provide insight into the development and current state of scientific knowledge in a given field.¹⁰ This approach allows the identification of trends that can guide future research and promote critical and constructive discussions.¹¹ In 2018, Digital Science & Research Solutions introduced Dimensions, an online platform designed to provide unique insights into research progress through the presentation of altmetric data.^12,13 The platform incorporates a range of metrics, including social media mentions, academic citations, clinical trials and commercial patents.¹³ While several bibliometric analyses have addressed topics related to cariology,^14–16 to date there has been no bibliometric and altmetric review focused on investigating the use of AI in this field. Therefore, this study aims to evaluate the use of AI in cariology through a bibliometric and altmetric review.

Methods

A bibliographic review was conducted with the aim of identifying studies employing AI in cariology. The search encompassed studies published until February 12, 2024, with no restrictions on language or year of publication. The bibliographic search was performed in the Web of Science – Core Collection (WoSCC) database, using the following search strategy: (ALL=(“Artificial intelligence” OR “Computational Intelligence” OR “Machine Intelligence” OR “Computer Reasoning” OR “AI-based” OR “Computer Vision Systems” OR “Knowledge Acquisition” OR “Knowledge Representation” OR “Machine learning” OR “Deep learning” OR “Expert systems” OR “Natural Language Processing” OR “neural networks”) AND (caries OR “dental caries” OR “Caries, Dental” OR “Dental Decay” OR “Carious Lesions” OR “Carious Lesion” OR “Lesion, Carious” OR “Lesions, Carious” OR “Decay, Dental” OR “Carious Dentin” OR “Carious Dentins” OR “Dentin, Carious” OR “Dentins, Carious” OR “Dental White Spot” OR “Spot, Dental White” OR “Spots, Dental White” OR “White Spot, Dental” OR “White Spots, Dental” OR “Dental White Spots” OR “white spot lesion*” OR “tooth caries” OR “teeth caries” OR “cariology”)).

Two researchers independently conducted the selection process, while the third author was consulted in case of discrepancies. Inclusion of studies occurred only when all reviewers reached a consensus on discrepancies. Publications were excluded based on the following criteria: not being research articles (conferences or editorials) and not including AI in cariology. Articles and reviews exploring the use of AI in cariology were considered for critical review and bibliometric analysis.

For each article, data such as number and density of citations, year of publication, journal, impact factor (2022) (Journal Citation Reports), study design, theme (main study objective and diagnostic method), country and continent, institution (based on corresponding author), authors, and keywords were collected. Study designs were categorized as: (1) proof of concept; (2) systematic reviews; (3) literature reviews; (4) observational; (5) randomized clinical trial. Considering the themes addressed in the studies, works were grouped according to the most prevalent themes, namely: (1) detection and diagnosis of dental caries; (2) prediction of dental caries; (3) classification of dental caries; (4) diagnosis and classification of dental caries; (5) diagnosis and prediction of dental caries. Regarding the diagnostic method utilized by AI, articles were grouped into: (1) radiographs, (2) computed tomography, (3) photographs, (4) transillumination; (5) biomarkers. Articles not addressing the diagnostic method were grouped as “not identified.”

Selected articles were entered into VOS Viewer software (version 1.6.18) (https://www.vosviewer.com/) to generate graphical representations highlighting collaboration among authors and keywords. Words associated with prominent focuses indicate higher occurrence, while words with the same colour and interconnected form networks that evidence more intense collaboration among studies. The correlation between the number of citations, year of publication, and journal impact factor was investigated using the SPSS statistical software for Windows (SPSS, version 24.0; IBM, Armonk USA Corp), with the Kolmogorov-Smirnov test to check the normality of data distribution and Spearman correlation test due to non-normal distribution. Alternatively, these statistical calculations (normality test and correlation) can be performed using Microsoft Office Excel, with the corresponding database available in the Data availability section.¹⁷ Altmetric data (performance of publications on social media, traditional media, and online reference managers) were measured using Dimensions (dimensions.ai).

Results

The bibliographic search identified a total of 355 records. For selection, titles, abstracts, and, when necessary, full texts were analysed. After this process, 175 articles were included for bibliometric analysis. Among the articles not included, five were editorials, 40 were conference papers, and the remainder (n=135) were excluded for not addressing the study’s objective.¹⁷

The selected articles accumulated a total of 2,760 citations in WoSCC. Of these, 884 were self-citations (32.02%). The average citations per year were 162.35. The most cited article in WoSCC was “Detection and diagnosis of dental caries using a deep learning-based convolutional neural network algorithm,” with 321 citations, published in the Journal of Dentistry.⁵ Spearman correlation revealed a very weak positive correlation between the number of citations and the journal impact factor (rho= ,195; p<0.05) and a strong negative correlation between the number of citations and the publication year (rho= -,734; p<0.001). The top 10 articles with the highest bibliometric performance in citation can be seen in Table 1.

The articles were published between 2008 and 2024, totalling 16 years since the first publication. The oldest ones date back to 2008 (n=2),^18,19 while the most recent ones, until February 2024, totalled four.^20–23 The year with the highest number of publications was 2023, with 57 articles, indicating a recent and significant interest in research related to this topic. The distribution of the number of publications over the years can be visualized in Figure 1.

Figure 1. Distribution of the number of publications over the years.

The most frequent journals that published articles on AI in cariology and their impact factors are presented in Table 2. Diagnostics (n=16) was the most prominent journal, followed by the Journal of Dentistry (n=12) and Journal of Dentistry Research (n=9). According to the Journal Citation Reports, the journals with the highest Impact Factors in 2022 linked to this study were NPJ Digital Medicine (IF 15.2), JAMA Network Open (IF 13.8), and Scientific Data (IF 9.8), all contributing with one article each.

Most articles were proof of concept (n=135), followed by literature review (n=17), systematic review (n=17), observational studies (n=3), and randomized clinical trials (n=3). Regarding the studies’ objectives, the following themes were identified: detection and diagnosis of dental caries (n=122), prediction of dental caries (n=30), classification of dental caries (n=13), diagnosis and prediction of dental caries (n=7), and diagnosis and classification of dental caries (n=3). Regarding the diagnostic method used, there was a highlight for the use of radiographs (n=99), followed by biomarkers (n=26), photography (n=22), transillumination (n=8), computed tomography (n=2), and “not identified” (n=18).

A total of 36 countries contributed to the articles related to AI in cariology. Considering the number of publications per country, the most prevalent ones were China (n=28), Germany (n=17), South Korea (n=14), and USA (n=14). Among the continents with the most articles, Asia stood out (n=96), followed by Europe (n=45) and North America (n=20). The worldwide distribution of publications can be observed in Figure 2.

Figure 2. Worldwide distribution of origin of publications on AI in cariology.

A total of 121 distinct institutions were identified among the corresponding authors of the selected studies. Charité University of Berlin (Germany) was the institution with the highest number of documents (n=13). Table 3 shows the top 10 institutions by number of documents. The authors with the highest number of articles using AI in cariology were: Schwendicke F (n=15), Krois J (n=13), and Orhan K (n=6). Figure 3 highlights collaboration and main author groups, while Table 4 shows the top 10 authors by number of publications.

Figure 3. Frequency and interaction of the main authors associated with the study.

A total of 597 keywords were identified. The most prevalent one was “artificial intelligence” (n=60), followed by “deep learning” (48 occurrences), “dental caries” (n=46), “machine learning” (n=45), and “diagnosis” (n=39). Figure 4 presents the most prevalent keywords (5 or more occurrences) and the collaboration relationships between them.

Figure 4. Frequency and interaction of the main keywords associated with the study.

According to Dimensions, 53 articles on AI in cariology have metrics for mentions across social media, traditional media, Mendeley, and patents. The highest-performing article was “The ADEPT study: a comparative study of dentists’ ability to detect enamel-only proximal caries in bitewing radiographs with and without the use of AssistDent artificial intelligence software” noted by 53 Mendeley users, 6 on X, 2 Facebook pages, and 2 news outlets.²⁴ Table 5 provides details for the top 10 articles with the highest altmetric attention.

Discussion

AI has been growing in prominence within dentistry, and this is reflected in its role in aiding numerous studies in cariology. Thus, this study was conducted to identify the main applications of AI in cariology through the analysis of the profile of publications involving this theme. It was observed from the results that this theme is mainly based on proof-of-concept studies, primarily seeking to detect and diagnose carious lesions, usually with the aid of radiographs. Most studies originated in China, and consequently, the Asian continent received greater prominence.

The most cited article identified in this bibliometric analysis was a proof-of-concept study published in 2018.⁵ This study aimed to evaluate the effectiveness of deep convolutional neural network algorithms for the detection and diagnosis of dental caries in periapical radiographs. The impressive metrics achieved by this study, in this short period of time, indicate the importance and current relevance of using AI in cariology. Despite the recent publication, this study was one of the first to investigate this theme, which may justify the high number of citations. Furthermore, dental caries is still the most prevalent disease in dentistry,⁶ which drives the number of studies linked to this disease, thus allowing a greater number of citations for relevant articles.

Self-citations refer to the act of an author citing their own works in their own scientific papers.²⁵ In contrast to other bibliometric analyses,^10,26 this analysis revealed a significant rate of self-citations. Although self-citations are not a common practice, in some cases, it becomes necessary to contextualize or expand on a previous work, that is, it can occur naturally due to research groups dedicated to a specific theme.²⁵

The oldest articles were published 16 years ago. These studies aimed at the detection and classification of dental caries through proof-of-concept. These studies showed promising results for the area of diagnosis in cariology. However, limited publications per year were observed until 2018. Only after 10 years, there was a continuous and increasing increase in publications on the use of AI in cariology, reaching 57 publications in 2023, which was the year of greatest prominence. This trend in the recent growth of publications indicates that this field has become more popular and has attracted increasing attention from researchers today.

The connection of AI with engineering and computer science fields was represented by the high number of journals with general or specific scopes in these areas (n=61). The most prominent journal was Diagnostics. Diagnostics (ISSN 2075-4418) is an international open-access academic journal on medical diagnostics. It publishes articles on molecular diagnostics, nuclear medicine, medical imaging, biomarkers, clinical decision support systems, and new diagnostic devices. Corroborating with the scope of the journal, it was observed that the studies published in this journal were mainly diagnostic and detection studies of dental caries. This journal also achieved a prominent position among the top 10 journals in another bibliometric analysis.²⁷

The most commonly employed study type was “proof of concept.” This research format is frequently adopted in the field of AI, representing a crucial step before the implementation of new technologies.^28,29 Moreover, it is considered a preliminary implementation aiming to assess whether a certain concept is feasible for practical exploration.²⁹ This characteristic may explain its widespread use as a study design. Nevertheless, it is worth mentioning that only a limited number of studies have performed clinical trials on the use of AI in cariology, as most of the existing literature consists of reviews. In light of this result, the need for more advanced research, including clinical trials, becomes evident, instead of relying exclusively on the determination of initial studies through proof of concept.

Dental caries is the most prevalent disease in dentistry worldwide. When considering initial lesions in clinical assessment, few individuals are unaffected.³⁰ Early diagnosis of this condition is crucial for preserving dental structure and providing appropriate treatment.⁸ Linked to the importance of diagnosing this condition, it was observed that this was the most researched theme in this study. Furthermore, the most used diagnostic method was radiographic imaging. In fact, radiography is the most accessible and recommended method to confirm suspected caries.³¹ AI was applied in most studies with the potential to process and segment radiographic images and identify caries lesions that were not identified by humans.

Regarding the objectives of the studies, the following themes were identified: detection and diagnosis of dental caries (n=122), prediction of dental caries (n=30), classification of dental caries (n=13). Regarding the diagnostic method used, there was a highlight for the use of radiographs (n=99), followed by biomarkers (n=26), photography (n=22), transillumination (n=8), computed tomography (n=2), and “unidentified” (n=18).

China emerged as the country with the highest volume of included articles. This nation also stood out in other bibliometric reviews addressing the use of AI in dentistry.^32,33 China demonstrates a vigorous internal stimulus for research advancement and technological innovation.³⁴ These results may indicate the predominance of the Asian continent. Oceania (n=1) and Africa (n=5) were the continents with the lowest number of identified articles. This result demonstrates the need for more research on this topic on these continents, especially in Oceania.

It was noted that a significant portion of the studies was conducted by various institutions, with Charite University of Berlin, based in Germany, standing out the most. This prominence can be attributed to author Schwendicke F, who played a prominent role in many of these studies, bringing visibility to the said institution. It is worth noting that this university also achieved a prominent position in another bibliometric analysis.³⁵

Schwendicke F was the author with the highest number of publications, and the research efforts of this author were primarily founded on proof-of-concept investigations into the detection and diagnosis of dental caries through radiographic examination. It is worth noting that this author collaborated on the only randomized clinical studies (n=3) published on the use of AI in cariology, and his publications were predominantly in the Journal of Dentistry. Next, Krois J stood out, who collaborated on studies published by Schwendicke F. In this way, this author also mainly published proof-of-concept investigations into the detection and diagnosis of dental caries. The third notable author was Orhan K, who also stood out by primarily publishing proof-of-concept for dental caries diagnosis.

Considering the investigated theme, the most frequently used keyword was “artificial intelligence.” In addition to its higher occurrence, this term revealed greater interaction with other keywords. In another bibliometric review, “artificial intelligence” also held a prominent position.³² Furthermore, still corroborating with the investigated theme, prominence was also observed for “dental caries.” As subcategories of AI, the terms “deep learning” and “machine learning” were also widely used. Associated with the most investigated theme “detection and diagnosis of dental caries,” another widely cited term was “diagnosis.”

A key benefit highlighted by this study is that the applied filters impose no limitations regarding time, citation, or language. This approach enabled a comprehensive and open analysis of all documents related to the topic published up to the date of the investigation. Although the choice to use a single database may be considered a limitation of this study, this decision was supported by other relevant bibliometric reviews.^10,11 Additionally, the bibliometric and altimetric data were manually validated by individual authors and recorded in a dataset.¹⁷

Conclusion

This bibliometric and altimetric review underscores the rising interest in AI applications in cariology, especially in Asia over the past five years. While most studies focus on AI’s potential to enhance diagnostic accuracy for dental caries, the scarcity of clinical trials highlights the need for validation in practical settings. Future research should prioritize real-world implementation to fully unlock AI’s impact on cariology and global oral health.

Ethics and consent

Ethical approval and consent were not required.