Deep learning-based segmentation of dental implants on cone-beam computed tomography images: A validation study

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Cone-beam computed tomography (CBCT) is crucial for dental implantology.
• Accurate segmentation of dental implants and prosthetic crowns is essential for treatment planning and assessment.
• Current segmentation methods can be time-consuming and prone to inaccuracies.

Purpose of the Study:

• To develop and validate a cloud-based convolutional neural network (CNN) model for automated segmentation (AS) of dental implants and prosthetic crowns.
• To assess the performance and efficiency of the AI model compared to traditional segmentation techniques.

Main Methods:

• A dataset of 280 CBCT scans was utilized, divided into training, validation, and testing sets.
• A CNN model was trained using expert semi-automated segmentation (SS) as ground truth.
• Performance was evaluated using timing, voxel-wise comparisons (Dice similarity coefficient), and 3D surface differences against refined-automated segmentation (R-AS).

Main Results:

• Automated segmentation (AS) was 60 times faster (<30 seconds) than semi-automated segmentation (SS).
• The CNN model achieved high Dice similarity scores for implant segmentation (0.92±0.02) and implant with restoration (0.91±0.03).
• Low root mean square deviation values (0.08±0.09 mm for implant only, 0.11±0.07 mm for implant+restoration) indicated high AI segmentation accuracy.

Conclusions:

• The cloud-based deep learning tool provides high-performance, time-efficient segmentation of dental implants on CBCT images.
• AI-based segmentation minimizes artifacts and enhances the creation of dental virtual models.
• Integration of this tool can improve pre-surgical planning and post-operative assessment in implant dentistry.