Enhancing Dental Diagnostics: Advanced Image Segmentation Models for Teeth Identification and Enumeration

Abstract

With recent advancements in Artificial Intelligence (AI) influencing various medical fields, dentistry faces several challenges. Among these challenges, accurate tooth counting and identification are essential for effective treatment and oral health monitoring. While several approaches exist for tooth identification and counting, they often entail drawbacks such as high costs or excessive manual labour. Panoramic X-ray imaging, a cost-effective and widely utilized method, is vital in dental healthcare, aiding in treatment planning and monitoring patient progress pre- and post-treatment. However, the complexity of panoramic X-rays, including non-uniform tooth shapes, misalignment, and overlapping teeth, pose challenges in tooth identification and counting. This study presents a novel approach to address these challenges by introducing a tooth identification and counting technique using advanced image segmentation models. We comprehensively evaluate multiple segmentation models, such as U-Net, Attention U-Net, Feedback U-Net, and Feedback U-Net with LSTM, specifically tailored to panoramic X-ray images, utilizing the open-source Tufts Dental Dataset. Our analysis demonstrates that the U-Net model surpasses other evaluated segmentation models for panoramic X-ray image segmentation because it can be effectively trained with limited datasets, which is crucial in dentistry where extensive labelled data is often unavailable. The primary goal of this research is to develop a technique that assists dental professionals in accurately identifying and counting teeth, thereby enhancing treatment planning and patient diagnosis. Code available on https://github.com/game-sys/Dental-Segementation-and-Enumeration.