Fast calibration and image reconstruction for magnetic particle imaging

Abstract

Magnetic particle imaging (MPI) is a relatively new medical imaging modality that images the spatial distribution of magnetic nanoparticles (MNPs) administered to the body. For image reconstruction with the system matrix (SM) approach, a time-consuming calibration scan is necessary, in which a single MNP sample is placed and scanned inside the full eld-of-view (FOV). Moreover, for a relatively large 3D high-resolution FOV, the reconstructed SM is too large to get high quality images in real-time using the standard state-of-the-art techniques. In this thesis, for the calibration scan, the use of coded calibration scenes (CCSs) is proposed, which utilizes MNP samples at multiple positions inside the FOV. The SM, which is sparse in the discrete cosine transform domain, is reconstructed using the Alternating Direction Method of Multipliers (ADMM) with l1-norm minimization. The e ectiveness of the CCSs for di erent parameter sets is analyzed via simulations, and the results are compared with the standard sparse reconstruction technique. As the MPI images are naturally sparse, ADMM is also proposed for image reconstruction, minimizing the total variation and l1-norm. Image quality is compared with the images obtained by widely used MPI image reconstruction algorithms: Algebraic Reconstruction Technique, Nonnegative Fused LASSO, and X-space-based projection reconstruction. Moreover, ADMM is parallelized on a GPU for real-time image reconstruction. The results show that the required number of measurements for system calibration is substantially reduced with the proposed methods, and the reconstruction performance is signi cantly improved in terms of both image quality and speed.