Detection of tooth decay and its severity based on processing of panoramic radiographic images

X-ray dental imaging helps dentists and radiologists diagnose dental disease and provide treatment plans for patients. In many cases, it is difficult to diagnose dental disease based on visual inspection alone. Therefore, the development of a support system for caries detection has been an issue in recent years. The most challenging type of image for this process is the Panoramic image. Because it includes other parts of the patient's mouth in addition to the teeth, the structures also have no clear boundaries. Here, we propose an automated detection system to detect tooth decay in full view images. The proposed model uses convolutional neural network to extract related features from X-ray. In this paper images are obtained from the Diagnostic Imaging Center of the Southwest State University of Bahia (UESB). 1,000 images of this dataset was manually tagged and 750 images were used as training data and the rest as test dataset. Our model achieved an accuracy score of 94% in the experimental set. Scores show acceptable diagnostic performance as well as speed up caries diagnosis, reducing the challenge of using a full-length photograph of real patients. Among the images with caries lesions in the experimental set, our model scored 64% and 69% for mild and severe cases, confirming the fact that severe caries spots are easier to diagnose and mild caries detection is more efficient. A stronger system is needed. This is a step towards developing an efficient, fully automated decision support system to assist experts in the field.