Harnessing Artificial Intelligence to Predict Dental Crowding in Modern Orthodontic Practice: A Comprehensive Review

Dental crowding is fundamental to orthodontic diagnosis and treatment planning, yet what constitutes "crowding" remains surprisingly heterogeneous. Space analysis quantifies tooth size-arch length discrepancy, irregularity indices characterize alignment displacement, and clinician-assigned severity reflects broader diagnostic interpretation. This narrative review examines the evolution of crowding assessment from conventional space analysis and digital measurement to two-dimensional artificial intelligence (AI), three-dimensional point-cloud learning, and clinician-aware agentic AI. Digitalization improves efficiency and reproducibility but does not resolve fundamental dependencies on how crowding is defined, anatomically represented, and established as a reference standard. Two-dimensional models enable prediction from readily available images but are constrained by projection and scale, whereas three-dimensional approaches directly learn from dental surface representations and can capture local geometric relationships. In a companion study of 495 arches from 284 patients, PointNet++ achieved an R² of 0.718 and a mean absolute error (MAE) of 1.54 mm for prediction of crowding. In a separate retrospective demonstration, a bounded Gemini-based Agent incorporating matched record-derived verbal information reduced MAE from 1.951 to 1.437 mm, whereas mismatched information degraded performance. These findings also illustrate that the target construct-the specific clinical feature of crowding that a model is trained to predict-may differ across assessment methods. Together, these findings highlight a fundamental challenge: improving prediction does not resolve ambiguity in the target construct. Future crowding AI should therefore align the target construct, reference standard, and input representation; preserve provenance and uncertainty; integrate complementary clinical information transparently; and route uncertain cases for expert review.