JMIR Publications

ABSTRACT

Background:

Early detection of tooth-related diseases plays a key role in dental health and the prevention of complications in patients. However, since dentists are not overly attentive toward tooth-related diseases that may be difficult to judge visually, there are cases where they miss treatment. The representative five tooth-related diseases (coronal caries or defect, proximal caries, cervical caries or abrasion, periapical radiolucency, and residual root) can be judged on panoramic images. To detect these, a web server was constructed to detect tooth-related diseases from the panoramic image input in real-time, which helped shorten the treatment planning time for scientists and reduce the misdiagnosis probability.

Objective:

This study designed a model to assess tooth-related diseases in panoramic images using AI in real-time. Therefore, this system can perform an auxiliary role in the diagnosis of tooth-related diseases by dentists and reduces the treatment planning time spent through telemedicine.

Methods:

In learning five tooth-related diseases, a total of 10,000 panoramic images were modeled: 4,206 coronal caries or defects, 4,478 proximal caries, 6,920 cervical caries or abrasion, 8,290 periapical radiolucency, and 1446 residual root. To learn the model, the Fast R-CNN, ResNet, and Inception models were used. Learning about five tooth-related diseases completely did not provide accurate information on the diseases because of the features of the panoramic picture being shadowy. Therefore, one detection model was applied to each tooth-related disease, and the model was integrated to increase accuracy.

Results:

In diagnosing five tooth-related diseases, the fast R-CNN model showed the highest accuracy, with an accuracy of over 90%. Therefore, the Fast-RCNN model was selected as the final judgment model and using it, the real-time dental disease diagnosis model can diagnose dental diseases that are difficult to judge visually by using the real-time tooth-related diseases detection model to assist dentists in treatment planning.

Conclusions:

Fast-RCNN showed the highest accuracy. Therefore, shortening the treatment planning time is possible by performing an auxiliary role in the dentist's diagnosis of tooth-related diseases using Fast-RCNN. In addition, by updating the captured panoramic images of patients on the web developed in this study, we are looking forward to increasing the accuracy of diagnosing five tooth-related diseases. for the five tooth-related diseases. Therefore, the dental diagnosis system of this study takes 2 min to diagnose five diseases in one panoramic image. It can judge dental information; therefore, it plays an effective role in setting a dental treatment schedule.