Answers are grounded in this paper and cite their sources.
This study aimed to preliminarily evaluate a Mask RCNN–Self Attention (SA) framework for automated CBCT segmentation with three-dimensional (3D) masks and volumetric assessment of lateral incisor root resorption (RR) associated with impacted maxillary canines. CBCT images from 42 patients with impacted maxillary canines were randomly divided into training, validation, and test sets (7:2:1) and used to train a Mask RCNN-SA framework. 10 bilaterally unaffected controls were retrospectively analyzed. A Mask RCNN-SA network was trained using sclices to segment the maxilla, impacted canine, lateral incisor, and first premolar. AI segmentation masks and volumetric measurements were compared with manual annotations, and manual segmentation was served as the reference standard. Root resorption severity was assessed using a voxel-based volumetric root-volume loss scheme. This volumetric grading was used to quantify root-volume loss. Segmentation performance, volumetric differences, exploratory SVM classification, and regression associations were assessed. The Mask RCNN-SA model achieved high segmentation performance (DSC: 93.60% for the maxilla and 92.50% for individual teeth; HD95 ≤ 2.0 mm). Automated segmentation was faster than manual segmentation. Volumetric analysis showed significantly smaller volumes of the maxilla and lateral incisors root on impacted sides compared with contralateral controls ( P < 0.01), however first premolar volume showed no significant difference. Voxel-based lateral incisor root volume (V) loss ranged from V0 to V4: V0 in 3 cases (7.1%), V1 in 16 cases (38.1%), V2 in 14 cases (33.3%), V3 in 8 cases (19.0%), and V4 in 1 case (2.4%). Most cases were classified as V1 or V2, although V3 and V4 categories were also observed. Impacted side status and sex were associated with lateral incisor root volume, but maxillary volume provided no additional explanatory power. In the exploratory association analysis, impacted side status and sex were associated with lateral incisor root volume. The exploratory SVM classifier model achieved an AUC but limited sensitivity for RR-positive cases. The proposed Mask RCNN-SA framework provided preliminary feasibility for automated CBCT segmentation and volumetric assessment of lateral incisor root associated with preliminary RR assessment.
No takes yet. Share an insight, caveat, or question.
Guo et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: