Deep learning-based 3D automatic segmentation of impacted canines in CBCT scans.

Area of Science:

• Dentomaxillofacial Radiology
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Impacted canines are common dental anomalies requiring precise localization for treatment planning.
• Manual segmentation of impacted canines on Cone Beam Computed Tomography (CBCT) scans is time-consuming and prone to variability.
• Deep learning offers a potential solution for automated and accurate segmentation.

Purpose of the Study:

• To develop and evaluate a deep learning model for automatic segmentation of impacted canines in CBCT scans.
• To assess the performance of the nnU-Net v2 architecture for this specific application.
• To compare the model's segmentation accuracy with established metrics.

Main Methods:

• Retrospective collection and annotation of 159 CBCT scans with impacted canines.
• Training a deep learning model using the nnU-Net v2 architecture.
• Performance evaluation using classification metrics (recall, precision) and segmentation metrics (Dice Similarity Coefficient, Hausdorff Distance, Intersection over Union).

Main Results:

• The nnU-Net v2 model achieved a recall of 0.90 and precision of 0.82 for impacted canine detection.
• Segmentation performance was strong, with a Dice Similarity Coefficient (DSC) of 0.84.
• Favorable results for 95% Hausdorff Distance (7.07 mm) and Intersection over Union (0.74) indicate good segmentation accuracy.

Conclusions:

• The nnU-Net v2 deep learning model demonstrates effective and autonomous segmentation of impacted canines in CBCT volumes.
• Artificial intelligence holds significant potential for enhancing diagnostic efficiency in dentomaxillofacial radiology.
• The developed model shows promise for clinical application in managing impacted canines.