一种基于树状结构关键点的弱监督血管分割方法
By using a loss function based on key points and a key point map correction annotation method, the problem of preserving shape and topology in tree structure segmentation is solved, achieving more accurate blood vessel segmentation, reducing annotation costs and improving segmentation results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING UNIV
- Filing Date
- 2025-08-05
- Publication Date
- 2026-07-17
AI Technical Summary
Existing automatic medical image segmentation methods struggle to effectively preserve the key geometric structures of target objects when dealing with tree-like structures, resulting in structural errors in the segmentation results, such as broken blood vessels and missing branches. Furthermore, the costly and labor-intensive annotation process makes accurate segmentation difficult to achieve.
We employ a loss function based on key points and a method to correct incomplete annotations using key point maps, including KP-Warp partial cross-entropy loss and KP-Warp Mumford Shah loss. Combined with the use of key point maps to correct annotations during iterative training, we improve segmentation performance under weakly supervised learning.
It achieves more accurate, efficient, and robust blood vessel segmentation in weakly supervised learning scenarios, reduces the workload of manual annotation, and improves the ability to preserve the shape and topology of the segmentation results, thus having high clinical application value.
Smart Images
Figure CN120953307B_ABST