Vit-Enhanced Ai Diagnosis Of Pediatric Malocclusion From Intraoral Images

This study aims to develop a multimodal artificial intelligence (AI) system "Oral iHub" based on Vision Transformer (ViT) to improve the efficiency and accuracy of malocclusion diagnosis in children. With ethical approval and parental consent, this study collected 1,500 three-view (left/frontal/right) Intraoral images from children aged 5-12 years between January 2024 and January 2025. After quality control, the valid dataset comprised 1,081 annotated samples with 10 types of malocclusion labels, plus 162 normal occlusion controls. A ViT-based framework was employed, integrating multi-view self-attention mechanisms and cross-entropy loss for multi-label optimization. Label randomization was applied to enhance robustness. The system's performance was comparatively evaluated against both GPT-4V analysis and blinded orthodontic specialist assessments. The AI system demonstrated robust diagnostic performance, generating interpretable clinical reports of pediatric malocclusion within 30 seconds of image submission. Quantitative analysis revealed: (1) 87.9% overall diagnostic accuracy (95% CI: 85.5–90.4%), (2) 79.2% precision in identifying complex malocclusions and (3) 86% agreement with orthodontic specialist evaluations. This innovative AI system "Oral iHub" was developed using mixed-dentition and early-permanent-dentition databases, achieving accurate pediatric malocclusion screening for early intervention. The technology demonstrates particular potential for improving orthodontic care accessibility in underserved regions.