Enhanced Tooth Region Detection Using Pretrained Deep Learning Models.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Imaging Analysis
• Deep Learning Applications

Background:

• Artificial intelligence (AI) is rapidly advancing healthcare technologies, including dentistry.
• Panoramic radiography and cone beam computed tomography (CBCT) are crucial for dental implant planning.
• Accurate identification of missing teeth positions is essential to minimize surgical risks.

Purpose of the Study:

• To develop a deep learning-based model for detecting missing teeth positions using segmented CBCT images.
• To evaluate the performance of various pretrained convolutional neural network (CNN) models for this task.
• To assess the impact of image segmentation on model performance.

Main Methods:

• Utilized a dataset of 500 CBCT images, divided into training (70%), validation (20%), and testing (10%) sets.
• Employed six pretrained CNN models: AlexNet, VGG16, VGG19, ResNet50, DenseNet169, and MobileNetV3.
• Compared model performance with and without applying image segmentation techniques.

Main Results:

• All tested deep learning (DL) models demonstrated high precision (above 0.90) for normal teeth classification.
• DenseNet169 achieved the highest precision (0.98), followed by MobileNetV3 (0.95), VGG19 (0.94), ResNet50 (0.94), VGG16 (0.93), and AlexNet (0.92).
• The DenseNet169 model showed strong performance in CBCT-based detection and classification, with 93.3% segmentation accuracy and 89% accuracy in classifying missing tooth regions.

Conclusions:

• Deep learning models, particularly DenseNet169, show significant promise for automated detection of missing teeth from CBCT images.
• The developed model can serve as a valuable, time-saving tool for dental implantologists.
• This represents a substantial advancement towards automated dental implant planning systems.