Deep learning-based system for automated staging of lower molar maturation.

Area of Science:

• Dentofacial Orthopedics
• Machine Learning in Medicine
• Radiographic Analysis

Background:

• Assessing patient maturation status is crucial for dentofacial orthopedic treatment planning.
• Dental development stages, per Demirjian's method, indicate skeletal maturity.
• Automating this assessment can enhance efficiency and orthodontic interventions.

Purpose of the Study:

• To develop and evaluate deep learning models for automated dental maturation stage assessment.
• To compare the performance of different convolutional neural network (CNN) architectures.
• To validate the models' clinical applicability in growth assessment.

Main Methods:

• A cross-sectional study analyzed 1805 panoramic radiographs.
• Four CNN architectures (Xception, ResNet, MobileNet, Inception) were trained on lower second and third molar maturation stages.
• Model performance was evaluated on combined, second molar only, and third molar only datasets; Grad-CAM visualized attention.

Main Results:

• The Inception model achieved the highest accuracy (0.96) on the combined dataset and (0.98) on the second molar dataset.
• ResNet performed best on the third molar dataset (accuracy 0.96).
• High inter-examiner agreement (kappa=0.94) and relevant structure focus confirmed by Grad-CAM.

Conclusions:

• Deep learning, particularly the Inception model, accurately classifies dental maturation stages.
• The system shows strong agreement with expert assessments.
• This AI tool can support clinical decision-making in orthodontic growth assessment.