Automated lesion detection in panoramic dental radiographs

Abstract

Even in the presence of physical indicators like pain, tumor, color, and function loss, determining the exact size or location of acute dental apical diseases is challenging. Even harder to detect is chronic apical periodontitis, which is asymptomatic. In such circumstances, use of dental radiography is especially beneficial. However, radiographs are not sufficient by themselves, and require interpretation by a well-trained dental specialist. Nevertheless, owing to the human factor, mistakes leading to incorrect treatment can be made by specialists because of a wrong diagnosis. This study aimed to introduce an automated dental apical lesion detection methodology by assessing changes in hard tissue structures. The system consists of modules for jaw separation, tooth segmentation, root localization, and lesion detection. Panoramic radiographs are used to improve the process of diagnosis. Unlike the column-sum methodology used in previous studies, the upper and lower jaws are separated using discrete wavelet transformation along with polynomial regression to obtain a better jaw separation curve. Subsequently, angular radial scanning is used to segment the teeth and capture the location of the tooth roots. At the last step, for each detected root, region growing is performed to detect possible lesions surrounding the root apices. The results for test samples indicate that use of the above-mentioned methods with proposed threshold selection is an effective way for discriminating anatomic structures from lesions, which is our main concern. The findings prove that the proposed methodology can be used efficiently as an assistant for examination of radiographs.