The accuracy of large language models in assessing tooth detection and identification, dental caries, periapical status, and tooth development from mixed dentition radiographs.
Researchers tested three large language models (GPT-5, Gemini 3 Pro, and Sonar) on their ability to detect teeth, identify caries, assess periapical status, and evaluate tooth development in radiographs of children with mixed dentition. Gemini 3 Pro performed best overall, but all models were less accurate than expert human assessment.
A plain-English summary of the study's own findings — we don't add our own claims or interpretation. Read the full paper for the complete methods and results, and apply your own clinical judgement.
Read the full study at the sourceRelated CPD from Vital Pulp
Once and for All: Does Flossing Prevent Caries?
What the evidence actually says — an honest, cited look at flossing, caries and gum health
The Recall — the month's most significant new dental research, summarised. Free.
Browse The Recall