Semi-supervised segmentation method based on motion-guided and dynamic nuclear magnetic resonance cine sequence of heart
By using a shared multi-scale feature encoding and dynamic convolution kernel iterative evolution method, combined with semantic upsampling guided by the correlation matrix, the problems of spatiotemporal consistency and annotation cost in cardiac magnetic resonance film sequence segmentation are solved, achieving high-precision and low-cost cardiac segmentation that adapts to the non-rigid deformation of the heart during contraction and relaxation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- FUJIAN PROVINCIAL HOSPITAL
- Filing Date
- 2026-04-20
- Publication Date
- 2026-07-17
AI Technical Summary
Existing cardiac magnetic resonance cinema sequence segmentation techniques suffer from insufficient spatiotemporal consistency, high cost of full sequence annotation, and boundary ambiguity caused by spatial heterogeneity when processing dynamic cardiac sequences, making it difficult to achieve high-precision, spatiotemporally coherent cardiac segmentation.
We adopt a shared multi-scale feature encoding architecture, combined with dynamic convolutional kernel iterative evolution with attention mechanism and semantic upsampling method guided by correlation matrix. Through motion-guided semi-supervised training framework, we achieve heart segmentation of the whole cardiac cycle with a small number of annotations, dynamically generate convolutional kernels that adapt to changes in heart morphology, and perform fine recovery guided by correlation matrix.
It achieves high-precision, spatiotemporally coherent heart segmentation throughout the entire cardiac cycle, significantly reducing the cost of manual annotation and improving the robustness and anatomical rationality of the segmentation results, especially in the segmentation accuracy of regions with blurred boundaries such as the apex of the heart.
Smart Images
Figure CN122415644A_ABST