Tooth Restoration and Dental Work Detection on Panoramic Dental Images via CNN

Abstract

Automatic detection of dental work and type of restorations plays an important role for human identification and creation of reports for dental treatment at clinics. In this study, we employed three state-of-the-art convolutional neural networks (CNNs), which are GoogleNet, DenseNet and ResNet, for classification of dental restorations. Implants, canal root treatments, amalgam and composite fillings, dental braces and unrestored teeth are the classes that are detected by the networks. The CNNs are validated on a dataset including 3013 tooth images. DenseNet has 94% accuracy which is the highest accuracy among three CNN architectures. Dental braces and implants are detected with more accuracy than other dental work.