F²CW-MPI: deep Wiener reconstruction method with fusion prior of system matrix–signal feature correlation in magnetic particle imaging

Objective. Magnetic particle imaging (MPI) provides high sensitivity and spatial resolution for visualizing magnetic nanoparticles, but its system matrix (SM)-based reconstruction remains an ill-posed inverse problem that is highly sensitive to noise interference. Existing iterative and learning-based approaches typically embed denoising priors at the image-domain level within the reconstruction process, which limits robustness to noise variation, relies on manually tuned parameters, and underuses the physical information embedded in the SM. Approach. In this study, we propose F2CW-MPI, a non-iterative deep Wiener reconstruction framework. Unlike existing methods that embed denoising priors at the image-domain level, F2CW-MPI moves noise suppression earlier into the signal domain. An adaptive feature-refined Wiener denoiser estimates noise characteristics from latent signal features and dynamically adjusts denoising parameters accordingly, enabling signal-dependent noise suppression. Building on this design, a cross-attention-based fusion module then models the correlation between the denoised signal features and the SM within the same feature space, embedding the SM’s physical prior into reconstruction. To reinforce this feature-level consistency, a joint loss function aligns the denoised signal features with ground truth features, together with image fidelity and concentration-distribution consistency terms. We evaluate F2CW-MPI on synthetic digital (training set: 6000 images; validation set: 2000 images; test set: 2000 images; with no overlap between sets) and vessel phantoms (training set: 80 subjects split into 28894 images; validation set: 10 subjects split into 2782 images; test set: 10 subjects split into 3078 images; with no overlap between sets), as well as real-world phantom data and in vivo data acquired on our homemade MPI scanner. Main results. Experiments show that F2CW-MPI outperforms iterative and existing learning-based methods in reconstruction accuracy and noise robustness, while preserving structural fidelity on real-world phantom and in vivo data. Significance. F2CW-MPI provides a robust and high-fidelity reconstruction strategy for MPI, showing potential for preclinical applications.

Authors

Institutions

Publication Details

Journal
Physics in Medicine and Biology
Published
2026-11-19
DOI
https://doi.org/10.1088/1361-6560/aec495
Primary Topic
Characterization and Applications of Magnetic Nanoparticles
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

F²CW-MPI: deep Wiener reconstruction method with fusion prior of system matrix–signal feature correlation in magnetic particle imaging

Lishuang Guo, Wenxuan Zou, Siao Lei, Zhen Dong et al.
Physics in Medicine and Biology
Characterization and Applications of Magnetic Nanoparticles
article

F²CW-MPI: deep Wiener reconstruction method with fusion prior of system matrix–signal feature correlation in magnetic particle imaging

Lishuang Guo, Wenxuan Zou, Siao Lei, Zhen Dong, Jiangang Liu, Yu An, Jie Tian, Jie He
article en

Abstract

Objective. Magnetic particle imaging (MPI) provides high sensitivity and spatial resolution for visualizing magnetic nanoparticles, but its system matrix (SM)-based reconstruction remains an ill-posed inverse problem that is highly sensitive to noise interference. Existing iterative and learning-based approaches typically embed denoising priors at the image-domain level within the reconstruction process, which limits robustness to noise variation, relies on manually tuned parameters, and underuses the physical information embedded in the SM. Approach. In this study, we propose F2CW-MPI, a non-iterative deep Wiener reconstruction framework. Unlike existing methods that embed denoising priors at the image-domain level, F2CW-MPI moves noise suppression earlier into the signal domain. An adaptive feature-refined Wiener denoiser estimates noise characteristics from latent signal features and dynamically adjusts denoising parameters accordingly, enabling signal-dependent noise suppression. Building on this design, a cross-attention-based fusion module then models the correlation between the denoised signal features and the SM within the same feature space, embedding the SM’s physical prior into reconstruction. To reinforce this feature-level consistency, a joint loss function aligns the denoised signal features with ground truth features, together with image fidelity and concentration-distribution consistency terms. We evaluate F2CW-MPI on synthetic digital (training set: 6000 images; validation set: 2000 images; test set: 2000 images; with no overlap between sets) and vessel phantoms (training set: 80 subjects split into 28894 images; validation set: 10 subjects split into 2782 images; test set: 10 subjects split into 3078 images; with no overlap between sets), as well as real-world phantom data and in vivo data acquired on our homemade MPI scanner. Main results. Experiments show that F2CW-MPI outperforms iterative and existing learning-based methods in reconstruction accuracy and noise robustness, while preserving structural fidelity on real-world phantom and in vivo data. Significance. F2CW-MPI provides a robust and high-fidelity reconstruction strategy for MPI, showing potential for preclinical applications.

Physics in Medicine and Biology
Beihang University (CN)
Characterization and Applications of Magnetic Nanoparticles
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon