Numbering tooth via a combination of YOLOv5 and tooth region information in RGB images for post mortem identification

Teeth are the best candidates for postmortem identification when there is a decay or severe condition in the human body. In recent years, many studies have utilized deep learning models for tooth detection and identification via X-ray images such as panoramic, bitewing and periapical images; however, studies involving RGB images are rare. DPR images require well-controlled clinical environments with specialized equipment. With the proliferation of mobile phones, there is an opportunity to use readily available visible light images in the preliminary diagnosis of certain conditions, which can save time and resources for medical treatment. The challenge in this study is the very small and limited dataset, which, if only using a deep learning model, will not have promising results. In this paper, we propose a combination of YOLOv5n and a postprocessing heuristic algorithm for tooth numbering in RGB color images. The results revealed that the proposed model achieved 50% more tooth numbering than did the individual deep learning model. This paper shows the possibility of using a custom heuristic algorithm to solve tooth RGB color image issues, and these results can also be used for edge or mobile computing due to the small amount of memory and computation.