Panoramic Radiographic X-Ray Image Tooth Root Segmentation Based on LeNet-5 Networks

Accurate teeth segmentation in panoramic radiographic X-Ray images is importance for orthodontic treatment and research. This paper evaluates the method of LeNet-5 convolutional neural network with input data of windowed image patches for automated tooth root segmentation. In total, 103,984 image patches created from 798 images are used for training and validation sets. The proposed method produced an accuracy of 87.94%, which is higher than comparative Sobel- and Canny-processed cases. A visual evaluation of the segmentation method shows a close resemblance to the ground truth. The method achieved high performance for automated tooth root segmentation on dental panoramic images. With some slight further modification and improvement, the proposed method might be applicable to be used in the first step of dental diagnosis or analysis systems, which involves similar segmentation tasks.