Panoramic radiograph quality assessment: Database and algorithm

Abstract

High-quality panoramic radiographs are crucial for providing accurate diagnosis and appropriate treatment to the patients, which have been widely employed in dentist clinic. Unfortunately, panoramic radiographs can be bothered with various distortions during the capture and imaging process, which can further negatively affect the judgment of dentists. Therefore, to deal with the challenge of panoramic radiographs quality assessment, we first carry out a large scale panoramic radiograph quality assessment database, which contains 2009 images labeled by experienced dentists. Then we propose a novel objective quality assessment method based on the convolutional neural network (CNN). Specifically, the multi-scale features from the different stages of the CNN are employed to make full use of both low-level and high-level information. Moreover, weight-sharing auxiliary networks are utilized to improve the quality understanding of the proposed method. The experimental results show that the proposed method outperforms all the compared methods for panoramic radiographs quality assessment tasks and is more capable of handling the practical medical diagnosis situations. The proposed method can also be used as preliminary screening tool and provide useful guidelines for the dentists.