International Journal on Magnetic Particle Imaging IJMPI
Vol. 11 No. 1 Suppl 1 (2025): Int J Mag Part Imag
https://doi.org/10.18416/IJMPI.2025.2503058

Proceedings Articles

Neural implicit representations for grid-agnostic MPI reconstructions

Main Article Content

Artyom Tsanda (Section for Biomedical Imaging, University Medical Center Hamburg-Eppendorf, Hamburg, Germany; Institute for Biomedical Imaging, Hamburg University of Technology, Hamburg, Germany), Sadia Khalid (Institute for Biomedical Imaging, Hamburg University of Technology, Hamburg, Germany), Martin Möddel (1) Section for Biomedical Imaging, University Medical Center Hamburg-Eppendorf, Hamburg, Germany; 2) Institute for Biomedical Imaging, Hamburg University of Technology, Hamburg, Germany), Tobias Knopp (1) Section for Biomedical Imaging, University Medical Center Hamburg-Eppendorf, Hamburg, Germany; 2) Institute for Biomedical Imaging, Hamburg University of Technology, Hamburg, Germany; 3) Fraunhofer Research Institution for Individualized and Cell-based Medical Engineering IMTE, Lübeck, Germany)

Copyright (c) 2025 Artyom Tsanda, Sadia Khalid, Martin Möddel, Tobias Knopp

Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 International License.

Abstract

Magnetic particle imaging (MPI) reconstructs the spatial distribution of magnetic nanoparticles on a fixed grid, the resolution of which is limited by the noise present in the system. This paper addresses the reconstruction problem while integrating single-image super-resolution for concentration maps.  We introduce Neural Implicit Representations (NIR) as an image prior, enabling arbitrary grid size sampling after training. Experimental results using a spiral phantom measurement reveal that NIR-based reconstruction maintains image sharpness across diverse grid sizes, surpassing the two-stage Kaczmarz-L2 reconstruction followed by bicubic up-sampling in preserving fine structural details. This technique has a potential for high-resolution MPI imaging without relying on extensive datasets.

Article Details