一种乳腺肿瘤图像多任务学习方法、系统、设备与介质
By improving the multi-task network and combining feature enhancement and fusion modules, the problem of insufficient feature fusion in breast ultrasound image segmentation and classification tasks is solved, achieving more accurate breast cancer image segmentation and classification, and providing an efficient automated assisted diagnosis solution.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ANQING NORMAL UNIV
- Filing Date
- 2025-11-07
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies for segmenting and classifying breast ultrasound images neglect the inherent relationship between low-level spatial features and high-level semantic features, resulting in insufficient fusion of low-level spatial features and high-level semantic features, which affects segmentation details and classification accuracy.
An improved multi-task network is adopted, which uses an encoder-decoder network with ResNet-18 as the backbone, combined with a hybrid convolutional (HC) module, a triaxial coordinate and channel collaborative attention (TACCSA) module, and a gated dynamic feature fusion (G-DFF) module to enhance feature fusion capabilities and achieve multi-task collaborative learning for segmentation and classification.
It enhances the ability to represent irregular shapes and complex textures, reduces the semantic gap, achieves more accurate breast cancer image segmentation and classification, and provides an efficient and accurate automated assisted diagnosis solution.
Smart Images
Figure CN121436076B_ABST