网织纤维染色病理图像智能分割方法及其应用
By constructing an appearance-structure dual-stream cross-attention network and a boundary-aware dynamic loss mechanism, the problem of insufficient explicit modeling of the topological destruction mode of woven fiber networks in existing technologies is solved, and accurate identification and efficient segmentation of the invasion boundary of pituitary neuroendocrine tumors are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN SHENGQIANG TECH
- Filing Date
- 2026-05-11
- Publication Date
- 2026-07-17
AI Technical Summary
Existing computer-aided analysis techniques cannot effectively model the topology of reticular fiber networks, resulting in insufficient accuracy in identifying the invasion boundaries of pituitary neuroendocrine tumors. Furthermore, existing segmentation methods cannot balance global continuity and microscopic boundary quality when processing ultra-high resolution images.
A dual-stream cross-attention network based on appearance and structure is constructed. By combining boundary-aware dynamic loss and counterintuitive graph model, and through multi-stage pre-classification filtering and physical scale-driven boundary awareness mechanism, the network can accurately capture fiber topological structure damage and finely optimize the boundary.
It improves the accuracy of tumor invasion boundary identification, generates segmentation results with high clinical sharpness and spatial closure, reduces computational costs, improves diagnostic efficiency, eliminates splicing artifacts, and provides reliable quantitative analysis basis.
Smart Images
Figure CN122176704B_ABST