基于非对称无监督扩散模型的CT转PET图像生成方法及系统
By decoupling the high- and low-frequency features of CT images using an asymmetric unsupervised diffusion model, and combining metabolic priors and topological consistency constraints, PET images that conform to physical laws are generated. This solves the artifact and topological structure problems in the CT-to-PET image conversion, improving image quality and clinical application value.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 南京市江宁医院
- Filing Date
- 2026-04-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies suffer from cross-modal frequency feature mismatch, distortion of physical imaging laws, and high-order topological structure breakage caused by local feature constraints during the process of generating CT images to PET images. This results in anatomical artifacts, background noise features that do not conform to biological laws, and risks to medical diagnosis in the generated PET images.
We employ an asymmetric unsupervised diffusion model, decouple the high- and low-frequency features of CT images through a frequency-domain adaptive structural encoder, construct a unidirectional diffusion generation process using metabolic prior-driven stochastic differential equations and topological consistency constraints, and perform joint unsupervised training with a lightweight proxy network to generate PET images that conform to the laws of physical imaging.
It eliminates the harsh, anatomical artifacts in generated PET images, improves the microscopic realism and topological consistency of the images, reduces computational resource requirements, and enhances the biological feature fit of the generated images, making it suitable for radiotherapy planning and non-invasive lesion screening.
Smart Images
Figure CN122415795A_ABST