A score-based model guided by signals for magnetic particle imaging reconstruction

Magnetic particle imaging (MPI) can reconstruct the distribution of magnetic nanoparticles (MNPs) from their nonlinear response signals. The traditional system matrix (SM) reconstruction method requires solving an ill-posed inverse problem using hand-crafted regularization terms that demand careful parameter tuning. Here, we proposed a score-based model guided by signals for MPI reconstruction. The measured response signals were embedded to guide the score-based model in sampling specific MPI images from noise. Simulated experiments showed that our proposed method improved the reconstruction quality in the presence of variable noise levels.