A Single-Stage Deep Learning Approach for Multiple Treatment and Diagnosis in Panoramic X-ray

This work investigates the recognition of multiple dental treatment and diagnosis conditions in a full scan dental panoramic image. In this study, we proposed a single-stage oriented deep learning model for five dental therapies, namely restoration, root canal treatment (RCT), RCT with crown, dental implant (DI), and DI with crown, two dental diagnoses (impacted tooth and tooth bud), and natural teeth without any treatment as tooth class. However, considering all eight classes to achieve a wide range of dental radiology data interpretation results in a highly imbalanced dataset. We considered a custom dataset of 661 dental panoramic images where a frequent category is tooth counted 11,282 times, while the least presence category is DI with the crown having 75 instances. The performance of the proposed work was evaluated with average precision of each treatment and diagnosis class. The result of RCT from the dental treatment category achieved an average precision score of 82.63% and recognition of impacted teeth with an average precision score of 18.78% from the dental diagnosis categories.