TaDA: tangent drawing agent for automatic gonial angle estimation on orthopantomogram

The gonial angle of the mandible measured in an Orthopantomogram (OPG) is being increasingly used in various dental treatments and research areas to assess and analyze individual characteristics of the subjects. The gonial angle is obtained by measuring the angle between the tangent lines drawn on the mandibular body and the ramus. Previous processes for this measurement were manually conducted, but there have been ongoing studies to automate this process. In the field of medical image analysis, many studies are being conducted to automate landmark search and line search tasks. Reinforcement Learning-based algorithms have shown higher search accuracy and precision compared to traditional search methods. As such, this study developed a Tangent Drawing Agent (TaDA) that can automatically draw tangent lines on OPGs using the Deep Q-Learning method, which is a type of reinforcement learning. TaDA draws a movable virtual line on the OPG and analyzes the image within a specific window size centered on the line, moving the line along the optimal path so that it becomes the tangent of the mandible. The proposed algorithm was compared and validated against existing search methods using OPGs of subjects, and the results showed that TaDA, constructed based on VGG16, could estimate the gonial angle with an accuracy of 0.57° error and a high search success rate of over 97.5%. The developed automatic TaDA is expected to be utilized in various medical image analysis fields for tangent search in the future.