Evaluation of automated tooth landmark localization on digital models

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate tooth landmark localization is crucial in digital dentistry.
• Manual landmark identification is time-consuming and prone to inter-examiner variability.
• Developing automated methods can improve efficiency and consistency.

Purpose of the Study:

• To evaluate the agreement between artificial intelligence (AI) and human examiners in tooth landmark localization.
• To assess the accuracy of an AI model in identifying specific tooth landmarks on 3D digital dental models.

Main Methods:

• A deep learning algorithm was developed to identify six landmarks per tooth on 3D digital dental models from 284 participants (5583 teeth).
• The AI model was trained and validated on separate datasets.
• Intra- and inter-examiner reliability was assessed by comparing human examiners' landmark identifications.

Main Results:

• The AI model demonstrated excellent intraclass correlation coefficients (0.97-1.0) with human examiners for all landmarks.
• AI localization error ranged from 0.01 mm to 1.68 mm.
• AI performed better than human inter-examiner differences for mesial and distal proximal points but had greater errors for cusp tip and facial axis points.

Conclusions:

• AI exhibits comparable tooth landmark localization accuracy to human examiners for measurements related to tooth size.
• Further improvements are needed for AI to match orthodontic clinicians' accuracy in identifying cusp tips and facial axis points.