Artificial Intelligence for Fast and Accurate 3-Dimensional Tooth Segmentation on Cone-beam Computed Tomography

IntroductionTooth segmentation on cone-beam computed tomographic (CBCT) imaging is a labor-intensive task considering the limited contrast resolution and potential disturbance by various artifacts. Fully automated tooth cannot be achieved merely relying CBCT intensity variations. This study aimed to develop validate an artificial intelligence (AI)-driven tool for imaging.MethodsA total of 433 Digital Imaging Communications in Medicine images single- double-rooted teeth randomly selected from 314 anonymized scans were imported manually segmented. An AI-driven algorithm based feature pyramid network was developed automatically detect segment teeth, replacing manual user contour placement. The evaluated volume comparison, intersection over union, Dice score coefficient, morphologic surface deviation, time.ResultsOverall, clinical reference segmentations resulted very similar volumes. mean union full-tooth 0.87 (±0.03) 0.88 semiautomated (SA) (clinical reference) versus fully (F-AI) refined (R-AI) segmentation, respectively. R-AI F-AI showed average median deviation SA 9.96 ?m (±59.33 ?m) 7.85 (±69.55 ?m), had time 6.6 minutes (±76.15 seconds), 0.5 (±8.64 seconds, 12 times faster), 1.2 (±33.02 6 faster).ConclusionsThis unique fast accurate approach imaging. These results may open doors applications surgical treatment planning oral health care.