Accuracy of Orthodontic Malocclusion Detection Using Multiple AI Models: A Comparative Study

Area of Science:

• Orthodontics
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate detection of orthodontic malocclusions is crucial for effective treatment planning.
• The application of artificial intelligence (AI) in analyzing medical images, including intraoral photographs, is an emerging field.
• Evaluating the diagnostic capabilities of general-purpose AI models in orthodontics is essential.

Purpose of the Study:

• To compare the accuracy of leading AI models (ChatGPT, Gemini, Claude, Copilot) in identifying orthodontic malocclusion features.
• To assess the agreement between AI model assessments and expert orthodontist evaluations.
• To determine the performance of AI in detecting specific malocclusion parameters from standardized intraoral photographs.

Main Methods:

• A cross-sectional study analyzed five standardized intraoral views from 50 children (aged 9-12).
• Eight malocclusion parameters were assessed: anterior crowding, diastema, overjet, overbite, molar/canine relationships, crossbite, and arch symmetry.
• Diagnostic accuracy was measured using Cohen's kappa and AUC, comparing AI models against an orthodontist's diagnosis.

Main Results:

• AI models showed poor to moderate agreement (Cohen's κ: -0.15 to 0.63) with orthodontist assessments.
• Easily visible features like crowding and diastema had higher accuracy (AUC: 0.56-0.85).
• Complex features requiring spatial analysis (overbite, crossbite, arch symmetry) had low agreement and near-random performance (AUC: 0.41-0.70).

Conclusions:

• Current multimodal AI models have limited, feature-dependent accuracy for orthodontic malocclusion detection.
• General-purpose AI systems are not yet sufficient for reliable orthodontic decision support.
• Development of specialized AI models trained on clinically annotated orthodontic datasets is necessary.