Automatic Tooth and Background Segmentation in Dental X-ray Using U-Net Convolution Network

Abstract

Tooth and background segmentation in dental X-ray is used to produce an area of a tooth by removing areas of tissue and other neighboring teeth. This presents challenges due to a large number of superimposed (overlapping) images of teeth between the adjacent teeth and the difficulty of determining the area of the tooth with other tissues automatically. This study proposes a new approach for the automatic segmentation of dental X-ray images using the U-Net convolution network. The stages used in the training process consist of data augmentation, pre-processing with Contrast Limited Adequate Histogram Equalization (CLAHE) and gamma adjustment, and training with the U-Net architecture. While the testing process consists of pre-processing, prediction, and removing small areas in the background. The experimental results show the average accuracy of the proposed U-Net convolutional network segmentation accuracy achieves excellent results, 97.61% compared to spatial Fuzzy C-means with gaussian kernel-based of 65.55%. It shows the proposed method achieves superior automatic tooth and background segmentation. The experiment result among 1907 image testing, there are 14.58% producing segmentation because of biased boundaries on the tissue at the root of the tooth and overlapping images on the enamel.