Weakly supervised intraoral photo detection and segmentation method for early caries
By generating high-quality pseudo-label masks through weakly supervised learning, the problems of high cost and insufficient segmentation precision of manual annotation in early caries detection are solved, and accurate detection and segmentation under different conditions are achieved, which has the value of automation and reproducibility in clinical applications.
Patent Information
- Application Number
- CN202512004841.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-29
- Publication Date
- 2026-06-05
- Estimated Expiration
- 2045-12-29
AI Technical Summary
Existing methods for detecting and segmenting dental caries rely on human experience, resulting in low detection and accuracy rates. In particular, they are difficult to achieve efficient large-scale screening in early dental caries detection. Furthermore, fully supervised learning methods suffer from high annotation costs and insufficient segmentation precision.
We employ a weakly supervised learning strategy, combining a small number of pixel-level annotations with a large number of image-level labels. We utilize a collaborative mechanism of fused activation heatmaps, two-layer superpixel grid expansion, and edge response maps to generate high-quality pseudo-label masks. We then train a segmentation network using partial cross-entropy loss, boundary intersection-union ratio loss, and consistency constraints to automatically generate tooth surface region masks and lesion masks.
It effectively reduces the cost of manual annotation, decreases false positives and false negatives, improves boundary fit and cross-sample stability, and achieves accurate detection and segmentation under different lighting, equipment and individual differences, and has automated and reproducible clinical application value.