一种基于跨模态一致性的单模态图像分割方法
By constructing a cross-modal consistent single-modal image segmentation method, and combining auxiliary and teacher networks with a contrast extraction network, the problem of difficulty in acquiring multimodal images and contrast differences in MR image segmentation is solved, achieving high-precision MR image segmentation and improving the accuracy of clinical diagnosis and treatment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SUZHOU UNIV
- Filing Date
- 2024-03-19
- Publication Date
- 2026-07-17
AI Technical Summary
Existing single-modal image segmentation methods have poor segmentation accuracy in clinical applications because patients cannot obtain multimodal images simultaneously, and the contrast difference in cross-sections of MR images is too large, which affects the diagnostic and treatment effects.
A single-modal image segmentation method based on cross-modal consistency is adopted. By constructing an auxiliary network, a teacher network, and a student network, combined with a contrast extraction network and multiple loss functions, the student network is trained to improve the segmentation accuracy of organs and tissues in MR images. Cross-modal consistency is measured using the Dice similarity entropy loss function and the Dice similarity contrast loss function, and the image contrast difference is reduced through a contrast alignment strategy.
The system effectively integrates two modalities of images during the training phase, and only requires a single modal image for segmentation during the inference phase. This significantly improves the segmentation accuracy of organs and tissues in MR images, reduces image contrast differences, and enhances the basis for diagnosis and treatment.
Smart Images
Figure CN118135222B_ABST