From Regression to Vision Transformers: A Narrative Review of Predictive Modelling in Dental Implantology and the Gap Between Algorithmic Performance and Clinical Adoption.

Predictive modelling has moved from simple regression equations to deep, image-native architectures capable of localising an implant position on a cone-beam computed tomography (CBCT) scan with sub-millimetre precision. Yet the discipline that builds these models and the clinicians who would use them appear, on current evidence, to occupy different timelines. This narrative review synthesises comparative performance data across traditional statistical, machine learning, and deep learning approaches to dental implant prognostication; situates these methods within the broader literature on prediction-model validation and reporting; and juxtaposes this technical trajectory against field survey data describing how practising dentists actually perceive and use predictive and digital tools. The synthesis suggests that while deep learning architectures now substantially outperform logistic regression and Cox models on discrimination metrics, routine clinical uptake remains constrained less by algorithmic ceiling and more by validation gaps, interpretability concerns, and infrastructural readiness at the chairside.