Intraoral scan segmentation typically combines geometric, intensity-based, and increasingly machine learning–driven methods to isolate individual teeth and soft tissues from a 3D mesh, but it is neither trivial nor fully reliable. Classical approaches rely on curvature and surface normal analysis to detect tooth boundaries at gingival margins and interproximal embrasures, often using region-growing or watershed algorithms from seed points on each tooth. These methods are particularly prone to errors interproximally, where tight contacts, overlaps, and noise can cause teeth to be merged or incorrectly split. Clustering and graph-based techniques can refine borders along sulci and contact points, but still struggle in areas of poor scan quality or undercuts. Even modern deep learning models that label each vertex or point as tooth or gingiva can misclassify at the contact points, so fully automated, error-free tooth-by-tooth segmentation for aligner design or restorative planning remains a significant technical challenge.