Abstract
Abstract
Objectives
The primary objective was to evaluate the predictive performance of artificial intelligence (AI) models for craniofacial and mandibular growth prediction. Secondary objectives were to assess methodological quality and identify research gaps.
Materials and methods
Nine databases were searched through August 12, 2026, supplemented by manual searches. Human studies assessing AI-based models for craniofacial or mandibular growth prediction were included. Study characteristics, prediction intervals, AI approaches, performance metrics, and validation strategies were extracted. Risk of bias and applicability were assessed using PROBAST.
Results
Twelve studies were included. Six assessed overall craniofacial growth, whereas six mandibular development. Prediction horizons ranged from ≤ 1.5 to approximately 10 years, with sample sizes of 33–639 participants. Predictive accuracy varied by anatomical region and methodological approach. Sex and malocclusion type were frequently used to define study populations but were not consistently incorporated as predictors, whereas ethnicity and skeletal maturation were rarely considered. Seven studies had overall high risk of bias and five unclear risk, mainly due to analytical limitations and inadequate validation.
Conclusion
AI-based models show promising predictive performance; however, current evidence is insufficient to establish superiority over traditional statistical methods. Overall high or unclear risk of bias and substantial heterogeneity in study populations and datasets, prediction horizons, outcome definitions, and AI modelling approaches limit confidence in the findings. Standardized multicenter datasets and independent external validation are required before clinical implementation.
Clinical relevance
AI-based growth prediction may support individualized orthodontic diagnosis and treatment timing, but current evidence does not support routine clinical use.