Canine impaction classification from panoramic dental radiographic images using deep learning models

Abstract

Maxillary canine impaction is a condition that commonly occurs in growing individuals with malocclusion, and identifying its type is always challenging for dentists when determining whether the patient needs a surgical intervention. To avoid severe complications, it should be detected and treated as early as possible. Classifying the type of the canine impaction from panoramic dental radiograph requires precise measurements and expensive training, which is time-consuming. The automation of this procedure could support dentists’ decision-making and save time and effort. Artificial Intelligence (AI) techniques including Machine Leaning (ML) and Deep Learning (DL) allow researchers to extract useful insights from data and thus improve decision-making. In this research, we apply DL technologies to classify impacted canines based on the Yamamoto classification. Four deep learning models were developed to classify the type of canine impaction from panoramic dental radiographic images: DenseNet-121, VGG-16, Inception V3, and ResNet-50. The results show that Inception V3 outperforms the other classifiers, with an accuracy of 0.9259. The proposed model could help dentists by automating the canine impaction prediction process, which not only reduce time and effort for dentists but also make the process easier and more reliable.