Spatial-Frequency Multi-Scale Transformer for Deblurring and Shape-Preserving Reconstruction in Magnetic Particle Imaging
Yaxin Shang, Jie Liu, Yanjun Liu, Yueqi Wang, Yusong Shen, Xiangjun Wu, Liwen Zhang, Hui Hui, Jie Tian
Abstract
Magnetic particle imaging (MPI) is a novel and emerging functional imaging technique that visualizes the spatial distribution of magnetic nanoparticles (MNPs). While the X-space method considers some important physical properties of MPI systems, it also neglects some phenomena, such as signals generated by MNPs outside (but close-to) the field-free region. Therefore, the X-space approach often results in blurring artifacts and incomplete edge information in native MPI images. In this study, we propose a spatial-frequency multi-scale transformer (SFM-Transformer) to address this limitation by restoring both the spatial and frequency domain features of the native image. SFM-Transformer comprises three modules: the spatial and frequency feature extractor module (SFFE), the spatial and frequency fusion module (SFF), and the multi-scale fusion module (MSF). By incorporating cross-feature space dependencies and capturing long-range details in spatial and frequency domains, our network captures pixel-level features and implicit physical properties features of native images. Furthermore, the SFM-Transformer utilizes a multi-scale strategy at the backbone to further improve performance. To facilitate comprehensive research, we construct a diverse dataset containing both simulated and experimental datasets. To validate the effectiveness of our method, we conduct extensive experiments in simulated and experimental data. The experimental results demonstrate that our method eliminates the blurring artifacts and recovers the edge shape of MPI images. This suggests that our approach has great potential for improving the accuracy and reliability of MPI for future applications.