Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Position-Prior-Guided Network for System Matrix Super-Resolution in Magnetic Particle Imaging

Xuqing Geng, Lei Su, Zhongwei Bian, Zewen Sun, Jiaxuan Wen, Jie Tian, Yang Du

TL;DR

This work tackles the costly SM calibration in Magnetic Particle Imaging by introducing Position-Prior-Guided Network (PPGnet), which embeds symmetric positional priors into a 3D RRDBNet-based SR framework to recover high-resolution SMs from undersampled data. The method learns a per-row mapping from low-resolution SM rows augmented with three coordinate channels to high-resolution SM rows, leveraging 3D convolutions and upsampling primitives. Across 2D and 3D OpenMPI data, PPGnet outperforms state-of-the-art methods (e.g., 3dSMRnet, TranSMS) in NRMSE for SM SR and improves image reconstruction metrics (PSNR, SSIM), particularly at higher undersampling. This approach accelerates SM calibration, reduces data collection burden, and improves MPI image quality, supporting faster and more feasible clinical deployment of MPI.

Abstract

Magnetic Particle Imaging (MPI) is a novel medical imaging modality. One of the established methods for MPI reconstruction is based on the System Matrix (SM). However, the calibration of the SM is often time-consuming and requires repeated measurements whenever the system parameters change. Current methodologies utilize deep learning-based super-resolution (SR) techniques to expedite SM calibration; nevertheless, these strategies do not fully exploit physical prior knowledge associated with the SM, such as symmetric positional priors. Consequently, we integrated positional priors into existing frameworks for SM calibration. Underpinned by theoretical justification, we empirically validated the efficacy of incorporating positional priors through experiments involving both 2D and 3D SM SR methods.

Position-Prior-Guided Network for System Matrix Super-Resolution in Magnetic Particle Imaging

TL;DR

This work tackles the costly SM calibration in Magnetic Particle Imaging by introducing Position-Prior-Guided Network (PPGnet), which embeds symmetric positional priors into a 3D RRDBNet-based SR framework to recover high-resolution SMs from undersampled data. The method learns a per-row mapping from low-resolution SM rows augmented with three coordinate channels to high-resolution SM rows, leveraging 3D convolutions and upsampling primitives. Across 2D and 3D OpenMPI data, PPGnet outperforms state-of-the-art methods (e.g., 3dSMRnet, TranSMS) in NRMSE for SM SR and improves image reconstruction metrics (PSNR, SSIM), particularly at higher undersampling. This approach accelerates SM calibration, reduces data collection burden, and improves MPI image quality, supporting faster and more feasible clinical deployment of MPI.

Abstract

Magnetic Particle Imaging (MPI) is a novel medical imaging modality. One of the established methods for MPI reconstruction is based on the System Matrix (SM). However, the calibration of the SM is often time-consuming and requires repeated measurements whenever the system parameters change. Current methodologies utilize deep learning-based super-resolution (SR) techniques to expedite SM calibration; nevertheless, these strategies do not fully exploit physical prior knowledge associated with the SM, such as symmetric positional priors. Consequently, we integrated positional priors into existing frameworks for SM calibration. Underpinned by theoretical justification, we empirically validated the efficacy of incorporating positional priors through experiments involving both 2D and 3D SM SR methods.

Paper Structure

This paper contains 26 sections, 17 equations, 5 figures, 5 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. METHOD
  4. Symmetry of SM with Respect to Position
  5. Signal Modeling
  6. 1D SM
  7. 2D SM
  8. Symmetry of SM
  9. Problem Statement
  10. Framework
  11. Position Information Embedding
  12. Feature Extraction
  13. Upsampling
  14. Experiment
  15. Experimental Setup
  16. ...and 11 more sections

Figures (5)

  • Figure 1: Architecture of the proposed PPGnet
  • Figure 2: The visual representation of three reconstructed 3D SM rows (central slice) for downsampling ratios of 2 (a) and 4 (b), respectively.
  • Figure 3: The visual representation of three reconstructed 2D SM rows (central slice) for downsampling ratios of 2 (a) and 4 (b), respectively.
  • Figure 4: The image reconstruction outcome for the shape phantom. The first row displays the reconstructed image, while the second row shows the corresponding 3D error map averaged along the Z-axis. The numbers “2” and “4” represent the downsampling ratios. The GT image is reconstructed using the measured full-size SM.
  • Figure 5: Visualization results of SM calibration from ablation study (2$\times$ sampling).