Tooth Instance Segmentation and Disease Detection With Uncertainty-Aware Contrastive Learning and Cross-Scale...

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Image Analysis
• Computer Vision

Background:

• Artificial intelligence (AI) is increasingly applied in dental treatments, with instance segmentation and disease detection as key research areas.
• Existing AI algorithms face challenges in accurately segmenting adjacent teeth and managing high model parameter counts.
• Unclear prediction boundaries and computationally intensive models hinder the clinical application of AI in dentistry.

Purpose of the Study:

• To propose UCL-Net, a novel lightweight AI framework for efficient tooth instance segmentation and disease detection.
• To overcome limitations of existing methods regarding boundary ambiguity and model complexity.
• To enhance the accuracy and efficiency of AI-driven dental diagnostics.

Main Methods:

• Utilizing uncertainty-aware contrastive learning with a multivariate Gaussian distribution for refined tooth segmentation boundaries.
• Developing a lightweight segmentation model with only 34.9 million parameters.
• Implementing cross-scale attention for efficient fusion of multi-scale features for improved disease detection.

Main Results:

• The proposed UCL-Net framework demonstrates significant improvements in tooth instance segmentation accuracy.
• The model achieves effective tooth disease detection with a reduced parameter count.
• Validation on four benchmark datasets confirms the lightweight and effective nature of UCL-Net.

Conclusions:

• UCL-Net offers a lightweight and effective solution for tooth instance segmentation and disease detection in AI-driven dental applications.
• The framework successfully refines segmentation boundaries and enables accurate diagnostics with a computationally efficient model.
• UCL-Net outperforms existing state-of-the-art methods, showing its clinical potential.