Fully automated method for three-dimensional segmentation and fine classification of mixed dentition in cone-beam...

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Image Analysis
• Deep Learning for Dental Applications

Background:

• Accurate classification and 3D segmentation of mixed dentition in cone-beam computed tomography (CBCT) images are crucial for diagnosing dentofacial deformities, supernumerary teeth, and artifacts.
• Existing methods face challenges in precision and automation, necessitating advanced analytical tools for improved diagnostic efficacy.

Purpose of the Study:

• To develop a high-precision, automated deep learning model for the fine classification and 3D segmentation of mixed dentition in CBCT images.
• To evaluate the model's diagnostic performance and compare it with human observers.

Main Methods:

• A deep learning model based on modified nnU-Net and U-Net architectures was developed for classifying and segmenting mixed dentition.
• The model was trained on 336 CBCT scans and validated on 120 mixed dentition and 143 permanent dentition CBCT scans from multiple centers.
• Performance was assessed using Dice similarity coefficient, Jaccard coefficient, precision, recall, F-1 score, and average symmetric surface distance, and compared against two human observers.

Main Results:

• The model achieved high accuracy in classification and segmentation for mixed dentition (Dice: 0.964, Jaccard: 0.931) and permanent dentition (Dice: >0.90, Jaccard: >0.90), even with fillings, malocclusion, or supernumerary teeth.
• Average symmetric surface distances were 0.091 ± 0.029 mm for mixed and 0.190 ± 0.092 mm for permanent dentition.
• The AI model significantly improved junior dentists' performance and increased segmentation speed by 20.9-22.8 times, while senior dentists showed no significant improvement.

Conclusions:

• The developed artificial intelligence model demonstrates strong clinical applicability, robustness, and generalizability for analyzing both mixed and permanent dentition in CBCT scans.
• This deep learning approach enhances diagnostic accuracy and efficacy, facilitating precise measurements for orthodontic planning and supporting dental education.