Segmentação tridimensional do canal mandibular com inteligência artificial

: estudo experimental

Abstract

Introduction: identification of the mandibular canal is critical for safe surgical planning in implant dentistry. Artificial Intelligence (AI) has been proposed as a tool for automated anatomical segmentation, but its accuracy requires validation. Objective: to evaluate the accuracy of three-dimensional mandibular canal segmentation using AI, compared with manual segmentation by junior dentists (JD) and a reference drawn by an experienced operator (gold standard – GS). Materials and Methods: ten cone-beam computed tomography (CBCT) scans were analyzed. Each scan was segmented manually by three junior dentists (JD), automatically by artificial intelligence (AI), and by a senior operator (gold standard – GS). The segmentations were converted into STL models and compared using Geomagic Control X software, based on the Root Mean Square (RMS) metric. Statistical analysis was performed using ANOVA and post hoc tests in SPSS software, version 28 (IBM Corp.), with a significance level of α = 0.05. Results: AI segmentation showed lower mean RMS values than manual segmentations on both mandibular sides (right: 0.7623; left: 0.8658) and demonstrated lower variability (reduced standard deviation). Statistically significant differences were found between AI and manual methods (p < 0.001), and among human operators. Conclusion: AI demonstrated higher precision and consistency in mandibular canal segmentation compared to JD, being closer to the reference (GS). AI may serve as a reliable clinical tool in implant planning, enhancing standardization and safety.

Date of Award	9 Jul 2025
Original language	Portuguese
Awarding Institution	Universidade Católica Portuguesa
Supervisor	André Correia (Supervisor) & Catarina Fonseca (Co-Supervisor)