Hierarchical Network Boosts 3D Dental Model Segmentation Accuracy

Posted: September 1, 2025
A novel U-shaped 3D dental model segmentation network has been developed to improve the accuracy of malocclusion segmentation in orthodontic diagnosis and treatment planning. The network utilizes a feature-guided deep encoder architecture with a normalization method and push-pull strategy to optimize point cloud density. By incorporating an inverted bottleneck global feature extraction flow, the network enhances semantic recognition of malformations. Experimental results show significant performance improvements over existing methods, achieving an overall accuracy of 96.6% and a mean intersection over union of 90.8%. The network's efficiency is demonstrated on self-constructed and public datasets, indicating its potential for advancing intelligent virtual orthodontics. This innovative approach addresses challenges in extracting complex geometric features, preserving key details in sparse regions, and avoiding confusion between neighboring category features, ultimately enhancing the precision of orthodontic treatment planning.
Views: 5
Sponsors
Townie Perks
Townie® Poll
Who or what do you turn to for most financial advice regarding your practice?
  
The Dentaltown Team, Farran Media Support
Phone: +1-480-445-9710
Email: support@farranmedia.com
©2025 Dentaltown, a division of Farran Media • All Rights Reserved
9633 S. 48th Street Suite 200 • Phoenix, AZ 85044 • Phone:+1-480-598-0001 • Fax:+1-480-598-3450