High-resolution magnetic particle imaging system matrix recovery using a vision transformer with residual feature network
🤖 AI Summary
In magnetic particle imaging (MPI), low reconstruction accuracy and high computational cost plague high-resolution system matrix recovery due to undersampling and non-uniform coil sensitivity. To address this, we propose a hybrid deep learning architecture integrating Vision Transformers (ViTs) with a multi-scale residual feature network. The ViT component captures long-range spatial dependencies via global self-attention, while the residual network enables robust feature fusion across scales, facilitating high-fidelity system matrix reconstruction from sparse measurements. An end-to-end training paradigm further reduces data acquisition requirements. Experiments demonstrate that our method preserves sub-millimeter spatial resolution while achieving a 23.6% improvement in signal-to-noise ratio (SNR) and accelerating reconstruction by 5.8× compared to conventional approaches. This offers an efficient, clinically viable pathway toward low-dose, rapid MPI.
Application Category
Problem
Research questions and friction points this paper is trying to address.
Recovering high-resolution system matrices in Magnetic Particle Imaging
Addressing resolution loss from downsampling and coil sensitivity variations
Enabling artifact-free system matrix recovery across multiple scales
Innovation
Methods, ideas, or system contributions that make the work stand out.
Hybrid deep learning combines transformer attention with residual networks
Dual-stage downsampling strategy simulates realistic MPI degradation conditions
Recovers both large-scale structures and fine details in system matrices
🔎 Similar Papers
No similar papers found.
