A CT image reconstruction method based on anisotropy and graph frequency dual domain fusion
By employing anisotropic Laplacian discretization, graph-guided multipath fusion, and frequency domain adaptive regularization modules, the problems of isotropic processing, limited multipath fusion strategies, and lack of frequency domain prior knowledge in low-dose CT reconstruction of the Mamba model were solved, achieving high-quality CT image reconstruction and improving the visual clarity and diagnostic value of the images.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANTONG UNIV
- Filing Date
- 2026-01-23
- Publication Date
- 2026-05-29
AI Technical Summary
Existing low-dose CT reconstruction methods based on the Mamba model have limitations in isotropic processing, a single multi-path fusion strategy, and a lack of prior knowledge in the frequency domain. These problems lead to blurred edges, loss of details, and a significant increase in noise in the reconstructed images, affecting diagnostic accuracy.
A CT image reconstruction method based on anisotropy and image-frequency dual-domain fusion is adopted. By combining anisotropic Laplacian discretization module, graph-guided multi-path state space modeling module and frequency domain adaptive regularization module with a complex feedforward network, a dynamic balance is achieved between directional adaptive feature capture, local structural integrity and global contextual relevance, and noise suppression and detail enhancement.
It significantly improves the visual clarity and diagnostic value of CT images. Through a modular collaborative architecture, it preserves key anatomical details while reducing noise, thereby improving the structural similarity and visual realism of the images, and has broad prospects for clinical application.
Smart Images
Figure CN122115644A_ABST