SISTR: Sinus and Inferior alveolar nerve Segmentation with Targeted Refinement on Cone Beam Computed Tomography images

Abstract

In dental implantology, precise delineation of maxillary sinuses and inferior alveolar nerves (IAN) on CBCT scans is essential for implant planning. Addressing the time-consuming manual segmentation, we introduce SISTR (Sinus and IAN Segmentation with Targeted Refinement), a novel deep-learning method for automated, precise segmentation. SISTR employs a two-stage approach: initially, it predicts coarse segmentation and offset maps to anatomical regions, followed by clustering for region centroids identification and targeted cropping for refined segmentation. Developed on the most diverse dataset to date for sinus and IAN segmentation, sourced from 11 dental clinics and 10 manufacturers (358 CBCT volumes for sinus, 499 for IAN), SISTR demonstrates robust generalizability. It achieved strong performance on an external test set, reaching average DICE scores of 96.64% (95.38-97.60) for sinus and 83.43% (80.96-85.63) for IAN, marking a significant 10 percentage point improvement in Dice Score for IAN compared to single-stage methods (p-value < 0.005). Chamfer distances of 0.38 (0.24-0.60) mm for sinus and 0.88 (0.58-1.27) mm for IAN affirm its precision. Efficient in fast and precise segmentation with an inference time of 4 seconds per case, SISTR advances implant planning in digital dentistry.