Deep Learning - Based System Matrix Restoration in Magnetic Particle Imaging
Artyom Tsanda, Sarah Reiss, Konrad Scheffler, Marija Boberg and Tobias Knopp from TUHH have shared their latest results on deep learning-based restoration of system matrices for Magnetic Particle Imaging (MPI).
System matrices are fundamental to MPI reconstruction and are typically measured using a small delta sample. However, measurement imperfections can introduce noise, low resolution, and missing or corrupted samples. While state of the art restoration methods in other domains rely heavily on deep learning, MPI system matrix data are scarce and lack ground truth, limiting the transferability of those methods.
To address this challenge, the authors leverage physics - based simulations to generate training data for system matrix restoration. The study demonstrates that the simulation to measurement domain shift is small enough for models to generalize effectively to real measurement data. The work covers four key restoration tasks: denoising, accelerated calibration, upsampling, and inpainting of corrupted system matrices.
Tsanda, Artyom; Reiss, Sarah; Scheffler, Konrad; Boberg, Marija; Knopp, Tobias (2026). Deep Learning for Restoring MPI System Matrices Using Simulated Training Data. Phys. Med. Biol. 71, 095029.
https://doi.org/10.1088/1361-6560/ae6016
