New AI method merges dental CT and scans with 0.55mm accuracy
GeDi+ICP outshines traditional methods with sub-millimeter precision in jawbone alignment
A new paper proposes a coarse-to-fine registration method combining a domain-generalizable local descriptor (GeDi) with the iterative closest point (ICP) algorithm to align jawbone CT and intraoral scanner (IOS) data. The study also introduces a pseudo-IOS evaluation framework that generates a point cloud emulating an intraoral scan from CT data, providing known ground truth for quantitative evaluation. In experiments on seven cases, GeDi+ICP maintained submillimeter mean absolute error (0.55–0.69 mm) across all evaluated jaw and artifact conditions, while ICP alone frequently converged to local minima from perturbed initial positions. Statistical analysis confirmed GeDi+ICP was significantly more accurate than GeDi alone, and metal artifacts had a statistically detectable but small overall impact.
- GeDi+ICP combines GeDi descriptors with ICP for 0.55–0.69mm alignment accuracy in jawbone CT/IOS registration
- Tested on 7 clinical cases; ICP alone failed to converge reliably, while GeDi+ICP maintained sub-mm precision even with metal artifacts
- Pseudo-IOS framework generates ground truth for rigorous evaluation, addressing a long-standing challenge in digital dentistry
Why It Matters
Enables precise pre-surgical planning and digital workflows in dentistry by fusing CT and intraoral scan data with sub-mm accuracy