Deep Learning Based Teeth Segmentation

This article discusses the importance of accurate teeth segmentation in dental diagnosis and treatment.Rapid advancements in Artificial Intelligence have led to the development of various approaches for example Res-UNet deep learning architecture.Res-UNet++ has been proposed as a refined version of the Res-UNet architecture to improve teeth segmentation performance.Res-UNet++ integrates three additional elements: squeeze and excitation block, atrous spatial pyramid pooling, and attention block.The purpose of these components is to enhance the performance of Res-UNet by improving the recalibration of features at both the channel and spatial levels, capturing multi-scale contextual information, and prioritizing the relevant regions of interest.Res-UNet++, UNet and Res-UNet were compared on two publicly available dental image datasets using evaluation criteria such as the dice coefficient and mean Intersection over Union (mIoU).The evaluation of these algorithms was implemented under the same experimental settings to statistically assess the significance of the enhancements.The result shows the superiority of Res-UNet++ over UNet and Res-UNet.The effectiveness of Res-UNet++ is demonstrated by its impressive assessment scores: the dice coefficient of 92.91% and 95.58% for the two databases, and the mean Intersection over Union (mIoU) of 88.68% and 88.72%.