Classification of Apical Openness Using Vision Transformer: A Comparative Approach with Expert Decisions

Area of Science:

• Dentistry
• Radiology
• Artificial Intelligence

Background:

• Root morphology evaluation is vital for dental diagnosis and treatment planning.
• Apical openness, indicating incomplete root development, poses challenges in endodontic and orthodontic treatments, particularly for young patients.
• Panoramic radiographs offer broad anatomical coverage with low radiation doses, making them suitable for dental assessments.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI)-based method for classifying apical root openness in panoramic radiographs.
• To assess the performance of a Vision Transformer (ViT) model in identifying different degrees of apical openness.

Main Methods:

• A dataset of 902 single-rooted permanent teeth from 512 panoramic radiographs was curated.
• Teeth were manually cropped and categorized into closed apex, anatomically open, and pathologically open groups.
• A Vision Transformer (ViT Base Patch32) model was employed for classification after image preprocessing.

Main Results:

• The ViT model achieved an overall accuracy, precision, recall, and F1-score of 88%.
• The AI model demonstrated more consistent classification outcomes compared to manual assessments by dental students.
• The model particularly outperformed less experienced dental professionals in classifying apical root openness.

Conclusions:

• The Vision Transformer (ViT) model shows high accuracy in detecting apical root openness on panoramic radiographs.
• This AI-based approach shows significant promise as a reliable tool for clinical decision support systems in dentistry.
• The AI method can aid dentists in diagnosis and treatment planning, especially in cases involving incomplete root development.