Multi-Slice Knowledge-Driven System Matrix Calibration in Magnetic Particle Imaging

Magnetic particle imaging (MPI) is a novel tomographic imaging technique with high sensitivity and high temporal resolution. Reconstruction methods based on the system matrix (SM) enable accurate estimation of the concentration distribution of magnetic nanoparticles. However, SM calibration measurement is highly time-consuming, and the SM needs to be recalibrated whenever the scan parameters, particle types, or even the particle environment change. Although previous studies have proposed methods to accelerate SM calibration, these approaches do not fully exploit the similarity between the two-dimensional (2D) SMs of adjacent slices. In this study, we propose a multi-slice knowledge-driven SM calibration method, MKD-SM, which leverages knowledge obtained from multiple adjacent x-y slices to improve SM calibration accuracy at high downsampling ratios. Specifically, based on the significant similarity of the 2D SMs from adjacent x-y slices, MKD-SM employs a cross-misaligned sampling method to obtain the low-resolution (LR) SM within the field of view (FOV), ensuring that the 2D LR SMs obtained from adjacent slices exhibit complementarity. Additionally, we use a federated affinity fusion method to aggregate the complementary knowledge across multiple adjacent slices and utilize an architecture based on a cascade of CNNs and transformers for high-resolution (HR) SM recovery. Experimental results on the public OpenMPI dataset demonstrate that MKD-SM outperforms existing calibration methods, achieving higher SM calibration accuracy, particularly at high downsampling ratios. Ablation studies confirm the effectiveness of leveraging knowledge from adjacent slices. Furthermore, the proposed method has been successfully applied to an in-house field-free line (FFL) MPI scanner, enabling HR image reconstruction with LR SM measurements.