Development of an AI Model for Predicting Skeletal and Soft Tissue Changes After Orthognathic Surgery: A Systematic Review

Abstract

Background: Accurate prediction of skeletal and soft tissue changes following orthognathic surgery is a critical prerequisite for effective preoperative planning, informed patient communication, and optimized surgical outcomes. Conventional prediction methods based on linear cephalometric ratios and commercially available 2D or 3D simulation software have demonstrated variable and often clinically insufficient accuracy, particularly in complex bimaxillary procedures and in predicting soft tissue responses to three-dimensional bone movements. Artificial intelligence (AI) and deep learning methods — including convolutional neural networks, graph neural networks, transformer architectures, and point-cloud-based models — have emerged as a transformative paradigm for orthognathic surgery outcome prediction, offering substantially improved accuracy, speed, and the capacity for joint skeletal and soft tissue forecasting within virtual surgical planning workflows.

Objective : To systematically review the available evidence on the development, validation, and clinical performance of AI models for predicting skeletal and soft tissue changes following orthognathic surgery, and to evaluate the current state and future directions of AI integration into virtual surgical planning in oral and maxillofacial surgery.

Methods: A systematic search of PubMed/MEDLINE, Embase, Scopus, Web of Science, IEEE Xplore, and the Cochrane Library was performed from database inception to December 2025, following PRISMA 2020 guidelines. Model development and validation studies, scoping reviews, systematic reviews, and primary clinical studies evaluating AI- or deep learning-based approaches to skeletal or soft tissue prediction after orthognathic surgery were eligible for inclusion. Data on AI architecture, imaging modality, prediction accuracy, prediction targets, and clinical applicability were extracted independently by two reviewers and synthesized narratively and in tabular form.

Results: Thirteen studies met the inclusion criteria, comprising five systematic or scoping reviews, one systematic review with meta-analysis, and seven primary model development, validation, or foundational accuracy studies, collectively addressing convolutional neural networks, graph neural networks, diffusion models, transformer architectures, deep biomechanical models, and 3D segmentation networks applied to orthognathic surgery prediction. AI models consistently outperformed conventional linear regression and simulation software approaches, with AI surpassing traditional linear regression in 6 of 32 evaluated prediction tasks in one scoping review and achieving accuracy ratings up to 98.7% in diagnostic AI models. Joint face-skeletal prediction models — simultaneously predicting both bone and soft tissue changes — demonstrated significantly reduced surface bias errors relative to single-modality approaches. Deep learning-based MRI segmentation achieved accuracy comparable to manual expert segmentation at approximately 400 times faster processing speed. Key limitations across the literature included small sample sizes, single-centre validation, and limited prospective clinical deployment.

Conclusion: AI-based models for predicting skeletal and soft tissue changes after orthognathic surgery demonstrate consistently superior accuracy to conventional methods, with emerging joint face-skeletal architectures representing a paradigm shift toward comprehensive, clinically actionable preoperative prediction. Despite this substantial methodological progress, the transition from research validation to routine clinical implementation remains limited by the absence of large-scale multicentre prospective studies and standardized validation protocols. Continued interdisciplinary collaboration between oral and maxillofacial surgeons, orthodontists, and computational imaging scientists is urgently required to accelerate this translational pathway.