The current study identifies and analyzes the top 100 most-cited articles pertaining to AI in orthodontics and orthognathic surgery, with the purpose of assisting future researchers in identifying emerging trends and areas that require further impactful research. Furthermore, this study acknowledged the authors, institutions, and sources that have made significant contributions to the advancement of AI in orthodontics and orthognathic surgery. The rationale behind combining orthodontic and orthognathic surgery in this study is their close association, as both fields can benefit from the application of AI as a valuable diagnostic tool.

As illustrated in Fig. 1, there is a remarkable exponential increase in AI publications from 2019 onwards. This growth aligns with McKinsey's Global Survey results, which revealed a significant rise in AI adoption from 20% in 2017 to 50% in 2022 [19]. While it is challenging to pinpoint the exact reason for the sudden surge of interest in AI research in orthodontics and orthognathic surgery, one possible explanation is the increasing adoption of AI technology in clinical practices. A similar trend was observed in the field of medical imaging, where the use of deep learning networks rapidly gained traction since its publication in 2016, peaking in 2020 [20].

Among the 100 most-cited articles, Orthodontics & Craniofacial Research emerged as the leading publication source, with the highest number of articles (n = 10). This achievement can be attributed to the journal's publication of a special issue on "Artificial intelligence and machine learning in orthodontics" in 2021, which had a significant impact on the orthodontics and orthognathic surgery communities. As a result, Orthodontics & Craniofacial Research surpassed Angle Orthodontist as the primary publication source for the top 100 AI articles. This observation underscores the potential influence of publishing special issue journals on research trends and demonstrates how such initiatives can help focus on emerging interests in the field and highlight new research applications.

According to our findings, six South Korean authors were among the top 10 contributors, and five South Korean institutions contributed significantly to the 100 most-cited AI articles. Notably, two of the top 10 authors, coming from non-orthodontic or orthognathic surgery backgrounds and affiliated with Relu BV, a dental software company specializing in AI algorithms for automating digital dental treatment planning, bring valuable perspectives from technological and engineering fields to advance our understanding of AI in this domain. Furthermore, South Korea leads with the highest number (n = 28) of publications among the 100 most-cited articles, closely followed by China (n = 16), Japan (n = 7), and United States (n = 7). These findings differ slightly from a previous study, which reported China as having the highest number of published studies on AI in dentistry, followed by the USA, South Korea, and India [21]. The differences in findings could be attributed to our analysis considering the country of origin of the corresponding author and focusing specifically on the field of orthodontics and orthognathic surgery, rather than encompassing the entire dental field. Moreover, Germany exhibited the highest citation-to-publication ratio among the top 100 cited AI articles, highlighting the impact and influence of their two publications. However, it is important to note that citation numbers solely reflect popularity and influence, and may not necessarily indicate the quality of the research [22]. Therefore, it is crucial to recognize as a limitation of the bibliometric study that the assessment of article quality was not conducted.

Unlike the country profile, which solely focuses on the nationality of the last corresponding author, the analysis of collaboration in our study encompasses the profiles of all co-authors. Our findings highlight that the most active international collaborations are observed between the USA, China, and South Korea. Furthermore, active collaborations are also observed between Brazil, Sweden, and Belgium. These variations in collaboration patterns between countries may be attributed to differences in research interests, funding resources, and languages, as suggested by previous studies [12].

AI can be categorized into two main types: narrow AI and strong AI. Narrow AI utilizes learning algorithms to solve specific tasks, and the knowledge acquired is not transferable to other tasks. On the other hand, strong AI refers to AI systems with human-level intelligence, possessing awareness and behavior similar to humans [23]. Strong AI aims to create a multi-task algorithm to make decisions in multiple fields. However, the development of strong AI raises ethical considerations and potential risks [24]. Currently, there are no strong AI applications in dentistry [25]. Table 5 displays the domains of the top 100 most-cited studies related to AI in the fields of orthodontics and orthognathic surgery. It is revealed that these studies primarily fall into categories such as automated imaging assessment (42%) and the application of AI in aiding diagnosis and treatment planning (33%). Cephalometric analysis, although an essential process in orthodontics, is prone to human error when performed manually [26, 27]. To address this, machine learning AI technologies such as convolutional neural networks have been developed for graphic image analysis. It utilizes multiple-layered connections to pass distinctive features to subsequent layers [28]. These advancements have facilitated the automation of cephalometric tracing and analysis, offering several benefits, including reduced human labor and decreased errors [29]. Popular keywords found in the studies included "deep learning," "machine learning," "convolutional neural network," and "automated identification,", highlighting the significant interest in these AI technologies. Notable examples of automated tracing and landmark identification systems, such as CephX (ORCA Dental AI, Israel) and WebCeph (AssembleCircle, South Korea) [30, 31]. Furthermore, AI algorithms can remove noise, enhance contrast and fine-tune images to provide dentists with clearer radiographs [32]. Regarding the accuracy of AI-facilitated cephalometric landmark detection, a systematic review conducted by Schwendicke et al. revealed high accuracy in detecting cephalometric landmarks in both 3D and 2D imaging [1]. However, there was notable heterogeneity in detection accuracy between individual landmarks. In 3D imaging, the proportion of landmarks detected within a 2 mm threshold was higher (0.870) compared to 2D imaging (0.792). Furthermore, a more recent study demonstrated significant accuracy in AI-facilitated 3D cephalometric landmarking, with a mean difference of 2.44 mm (95% CI 1.83–3.05) between automated and manual landmarking [33]. Interestingly, such discrepancy showed a decreasing trend over the years, suggesting advancements and improvements in AI technology.

Furthermore, it is noteworthy that the training and testing sample sizes varied significantly among the different studies, ranging from 18 to 20480 for the training sample size and 6 to 5120 for the testing sample size. Interestingly, our analysis identified a significant positive correlation between the testing sample sizes and the citation counts indicating that papers with larger testing samples tend to receive more citations. Furthermore, earlier publication dates were associated with higher citation counts. However, it is essential to consider that other factors, such as the research topic, journal, authors, and institutions, may also influence the citation counts.

One limitation of this study is the relatively recent emergence of research on the application of AI in orthodontics and orthognathic surgery, which may contribute to a lower number of citations compared to more established dental topics with a longer research history. Additionally, while the authors' profiles were analyzed to explore collaboration between countries, the specific level of contribution from each author could not be determined, potentially leading to an overestimation of collaboration. Furthermore, the quality of the included studies was not sufficiently assessed, and the level of evidence may be varied. Lastly, the use of the WoS Core Collection database, which primarily includes English-language articles, may have resulted in the exclusion of impactful studies published in other languages.