A magnetic nanoparticle imaging reconstruction method based on a cascaded neural network
The magnetic nanoparticle imaging method using a two-stage cascaded neural network and a dual-loss training strategy solves the problems of insufficient high spatial resolution and quantitative capability in existing technologies, and achieves high-resolution, low-artifact magnetic nanoparticle imaging reconstruction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2026-05-15
- Publication Date
- 2026-07-17
AI Technical Summary
Existing magnetic nanoparticle imaging technology struggles to balance high spatial resolution with good quantitative capabilities during image reconstruction and is sensitive to noise, resulting in insufficient quantitative accuracy and stability.
A two-stage cascaded neural network and a dual-loss training strategy are adopted. The global and local features of magnetic nanoparticles are processed by feature extraction subnetwork and feature enhancement subnetwork, respectively. By utilizing the nonlinear magnetization response information of magnetic nanoparticles, combined with the encoder-decoder U-Net structure and spatial attention module, the staged reconstruction of images is achieved.
It improves the spatial resolution and quantitative accuracy of magnetic nanoparticle imaging, reduces artifacts, achieves a balance between high resolution and good quantitative capability, and enhances the visual effect and reliability of reconstructed images.
Smart Images
Figure CN122409446A_ABST