Artificial intelligence (AI), machine learning (ML), and deep learning (DL) are revolutionizing orthognathic surgery and maxillofacial traumatology by enhancing diagnostic imaging and three-dimensional virtual surgical planning (VSP). This study provides a comprehensive bibliometric analysis of the field, outlining its thematic structure, temporal development, international collaboration patterns, and the clinical limitations documented in the literature. A systematic search across Scopus, PubMed/MEDLINE, and IEEE Xplore identified 854 eligible records. Keyword co-occurrence analysis revealed five thematic clusters, while temporal analysis highlighted generative AI and large language models as the newest research fronts. Country-level co-authorship showed dominant contributions from China and the United States, with the latter acting as a central connector. Key challenges to clinical translation include methodological heterogeneity, dataset homogeneity, and limited algorithmic interpretability.
Introduction and Background
Artificial intelligence, machine learning, and deep learning are fundamentally transforming orthognathic surgery and maxillofacial traumatology. The shift from traditional two-dimensional planning to three-dimensional virtual surgical planning (VSP) has paved the way for algorithmic automation, enabling AI to enhance radiographic diagnosis, improve VSP, predict treatment outcomes, and streamline patient care. Despite these advancements, challenges such as methodological inconsistency, dataset limitations, and ethical concerns hinder widespread clinical integration. This bibliometric analysis aims to quantitatively map the scientific output, identify research trends, and understand collaboration structures in this evolving specialty.
Methodology for Bibliometric Analysis
A systematic search was conducted across Scopus, PubMed/MEDLINE, and IEEE Xplore from their inception until August 24, 2026, restricted to English-language records. The search query combined AI-related terms with craniomaxillofacial and surgical terms. Records were included if they implemented an AI method and applied it to the craniomaxillofacial or dental domain, including educational and patient communication aspects. Exclusion criteria included editorials, conference proceedings volumes (unless individual papers), studies without an AI method, head-and-neck oncology, non-surgical facial reconstruction, and non-clinical computer vision. An initial search retrieved 1,925 records, which were reduced to 854 unique eligible documents after duplicate removal and two independent AI-assisted screening passes with author adjudication. Data extraction focused on metadata like title, authors, publication year, keywords, and affiliations, with bibliometric networks constructed using VOSviewer for co-occurrence and co-authorship analysis.
Thematic Domains and Research Hotspots
Keyword co-occurrence analysis identified five thematic clusters reflecting the interdisciplinary nature of the field. Cluster 1 focuses on Generative AI and Large Language Models, primarily for educational applications, clinical decision support, and patient communication. Cluster 2 encompasses Deep Learning and Computer Vision, addressing network architectures, 3D imaging, segmentation, and virtual surgical planning. Cluster 3 centers on Dentomaxillary Radiological Imaging, with applications in mandibular canal segmentation and third molar risk assessment. Cluster 4 represents Orthognathic Surgery and Cephalometric Analysis, using ML/CNN for treatment planning and malocclusion classification. Finally, Cluster 5 links Computer-assisted and Reconstructive Surgery with virtual/augmented reality planning, treatment outcome assessment, and personalized 3D-printed solutions for conditions like temporomandibular joint disorders.
Temporal Evolution and Global Collaboration
The field is experiencing rapid growth, with annual publication output significantly increasing since 2020, and generative AI applications emerging as the most recent thematic front compared to more established areas like cephalometric automation. The international co-authorship network revealed 35 significant contributing countries, structured into six regional clusters. China and the United States are the leading contributors by publication volume (163 and 155 documents, respectively), while the United States holds the most central position, connecting diverse research communities and accumulating the highest citation count (2,404 citations). This highlights a partly regional structure of collaboration with key connector countries.
Clinical and Technical Barriers
Several barriers consistently impede the clinical translation of AI in orthognathic surgery. Methodological heterogeneity stemming from varied imaging protocols (e.g., 2D cephalograms vs. 3D CBCT) and annotation differences limits cross-study compatibility and model accuracy. Dataset bias is prevalent, including single-center bias, demographic bias, spectrum bias (over-representing clear-cut cases), and class imbalance. Algorithmic interpretability remains a significant 'black-box' challenge for deep learning networks, posing ethical and medicolegal concerns in a field where errors can lead to permanent sequelae. Specific risks of generative AI include high rates of fabricated references ('hallucinations') and a lack of formal training for residents on model limitations and safety.
Future Strategic Directions and Limitations
Future progress can be achieved by developing multimodal fusion architectures that integrate diverse imaging data, adopting federated learning frameworks to allow cross-institutional model training without raw patient data transfer, and implementing explainable AI methods (e.g., gradient-weighted class activation mapping) to build clinical trust and maintain the surgeon's central role in decision-making. The study acknowledges limitations, including its restriction to English-language records from three databases, reliance on AI-assisted screening with moderate agreement, co-authorship analysis limited to the Scopus subset, and potential inconsistencies in keyword terminology. It emphasizes that bibliometric mapping describes literature structure and does not assess clinical validity or safety.
Conclusions
AI research in orthognathic and maxillofacial surgery is characterized by five evolving thematic clusters, with generative language applications emerging as a significant new frontier alongside established work in cephalometric automation and surgical planning. International collaboration shows a regional structure, led by China and the United States. Despite rapid advancements, widespread clinical adoption is hindered by persistent challenges such as methodological heterogeneity, dataset homogeneity, and limited algorithmic interpretability. Addressing these barriers through multimodal architectures, federated learning, and explainable AI is crucial for future clinical translation, ensuring the maxillofacial surgeon remains central to the decision-making process.
