A review of tooth AI segmentation on medical data

Tooth automatic segmentation on medical data is a critical prerequisite for ensuring accurate diagnosis, effective treatment planning, and advancing digital dental healthcare practices. The continuous evolution of deep learning technologies has led to the widespread adoption of various neural network architectures–such as U-Net, Mask R-CNN, Vision Transformer, and the recently proposed Mamba framework–significantly enhancing segmentation performance in both 2D and 3D modalities. Initially, we offer a comprehensive overview of automatic dental segmentation research based on dental medical images, analyzed using CiteSpace software. Subsequently, we summarize and incorporate research findings published between 2016 and 2025, which include 11 mainstream open-source datasets spanning various imaging modalities such as Cone Beam Computed Tomography (CBCT), Dental Panoramic Radiographs (DPRs), and Intraoral Scanners (IOS). A total of 72 representative studies are included, systematically categorized according to foundational network architectures, 2D segmentation methods, and 3D segmentation techniques. We provide an in-depth analysis of the advantages, limitations, and performance evaluation metrics associated with each segmentation method. Although heterogeneity resulting from varying datasets, annotation standards, and the absence of external validation restricts the comparability of research outcomes, the establishment of refined standards is gradually positioning automatic dental segmentation as a vital component of digital dentistry.