Predicting jaw growth with AI: results still insufficient for clinical practice

Twelve studies, very different predictions

The researchers searched nine databases up to August 12, 2026, then supplemented this with a manual search. Six studies focused on overall craniofacial growth and six on development of the mandible, the lower jaw.

Sample sizes ranged from 33 to 639 participants. Prediction horizons ranged from a year and a half or less to approximately ten years. These differences matter: predicting short-term changes and anticipating several years of growth are not the same task.

Accuracy varied by anatomical region and the method used. The abstract consulted does not provide a numerical measure that would allow this performance to be translated into a margin of error usable for an individual patient.

Validation is the main obstacle

The methodological assessment using the PROBAST tool classifies seven studies as having a high overall risk of bias and five as having an unclear risk. The problems mainly concern the analyses and insufficient model validation.

The populations and data used also differ considerably, as do prediction horizons, measured outcomes and AI approaches. This heterogeneity limits confidence in the reported performance and comparisons between studies.

Sex and type of malocclusion were frequently used to define the study populations, without being systematically included as predictive variables. Ethnicity and skeletal maturation were rarely taken into account. The review thus identifies gaps in how differences between patients are represented.

A promising avenue for orthodontics, not yet a routine tool

The potential benefit is to help personalize orthodontic diagnosis and treatment timing. But promising predictive performance is not enough to demonstrate that a model outperforms a traditional statistical method.

The authors call for standardized multicenter datasets and independent external validation before clinical implementation. The aim is to verify that predictions remain reliable on data other than those used to develop the models.

What we can take away

AI could help anticipate craniofacial growth, but this review supports neither its superiority over conventional methods nor its routine clinical use.