List of Papers Browse by Subject Areas Author List
Abstract
Magnetic Particle Imaging (MPI) is a promising tomographic medical imaging. However, System-Matrix (SM) calibration in MPI is prohibitively time-consuming. As a result, MPI image reconstruction using a sparse SM has attracted increasing attention. Nevertheless, reconstructions with a sparse SM often suffer from degraded in-plane details and structural discontinuities along the axial direction. To this end, we propose the BiSA-DiT framework, which improves tomographic reconstruction quality by alternately applying a powerful spatially aware generative prior with a physics-driven data consistency module. Specifically, by incorporating bidirectional synergistic Axial and Planar Spatial Transformer blocks, the generative-prior operator captures rich in-plane details while maintaining robust axial consistency. During the iterative optimization process, we employ an ADMM solver to ensure that the reconstructed results remain consistent with the MPI physical forward model. We validated BiSA-DiT on clinical coronary
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1442_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{ZhaXun_BiSADiT_MICCAI2026,
author = { Zhang, Xun AND Zou, Wenxuan AND Zhou, Yixiang AND Chen, Ziwei AND Zhang, Zeyu AND Tian, Jie},
title = { { BiSA-DiT: A Bidirectional Spatially-Aware Diffusion Transformer for Sparse Tomographic Magnetic Particle Imaging } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16888},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes the BiSA-DiT framework to address the degradation of in-plane details and axial structural discontinuities in MPI reconstruction caused by sparse system matrix. The authors introduce a bidirectional spatially-aware generative prior operator, which incorporates parallel PST and AST blocks to provide conditional guidance for the DiT-based denoising process. This architecture enables the synergistic capture of fine in-plane details and robust axial consistency. By alternately applying this generative prior with an ADMM-based physics-driven data consistency module, BiSA-DiT significantly outperforms current SOTA methods in coronary MPI reconstruction.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.Rather than adopting a simplistic end-to-end black-box mapping, the framework integrates the generative prior with the MPI physical forward model through an ADMM solver. This decoupled and alternating strategy not only enhances the model’s capability to handle undersampled data but also strengthens the physical fidelity of the reconstructed results, providing robust interpretability. 2.The authors designed parallel PST and AST blocks to effectively extract spatial features, which are fed into the DiT backbone as conditional guidance. This architecture enables the generated priors to possess both in-plane details and axial consistency. 3.Under 4x sparse System Matrix conditions, Table 1 shows that BiSA-DiT achieves significant improvements in PSNR, SSIM, and nRMSE across 6 baselines. The low standard deviations observed across various cases demonstrate the model’s exceptional robustness and stability. Qualitative results in Figure 2 further show that the 3D MPI vascular images reconstructed by BiSA-DiT maintain high-fidelity details and greatcontinuity. These results suggest the framework holds great promise for fast and highquality vascular imaging in real-world MPI applications. 4.The ablation study validates the essential roles of both PST and AST modules. Table 2 indicates that while each module individually surpasses the “Pure DiT” model, their integration yields the best results, successfully justifying the dual-branch spatial awareness design.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.In Section 2.3, the authors describe a ‘one-step denoising’ process based on Tweedie’s formula; however, this phrasing presents terminological ambiguity. Within the diffusion model literature, ‘single-step denoising’ typically refers to methods that complete reconstruction via a single forward pass. This description may mislead readers into assuming that the entire reverse sampling trajectory is finished in a single step, thereby obscuring the iterative nature inherent to diffusion models when addressing ill-posed inverse problems. 2.Furthermore, the manuscript lacks technical specificity regarding the implementation and formulation of the diffusion framework. Although a DDPM-based approach is cited, the authors do not clarify whether accelerated sampling strategies, such as DDIM [1], were utilized during the inference phase. This omission hinders an objective evaluation of computational efficiency, which is an important consideration for the real-time clinical application of MPI. 3.Equation (3) utilizes S^T for the system matrix, however, it should be clarified whether only real-valued components were used, as the appropriate operator for complex-valued matrix is typically the conjugate transpose S^H. 4.The manuscript does not specify whether the underlying diffusion model is based on a Variance Preserving or Variance Exploding formulation. Since these two frameworks involve different noise schedules and reverse-time SDE solvers [2], this omission may affect the reproducibility of the proposed method. [1] Song, Jiaming, Chenlin Meng, and Stefano Ermon. “Denoising diffusion implicit models. “ arXiv preprint arXiv:2010.02502 (2020). [2] Karras, Tero, et al. “Elucidating the design space of diffusion-based generative models. “ Advances in neural information processing systems 35 (2022): 26565-26577.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
My recommendation is based on the following major factors: 1.Synergy between Physical Model and Generative Prior: A key strength of this work is that it moves beyond simplistic “black-box” mappings. By utilizing an ADMM solver, the framework organically integrates a generative prior with the MPI physical forward model. This decoupled and alternating strategy not only enhances the model’s capability to handle highly undersampled data but also significantly strengthens the physical fidelity and interpretability of the reconstructed results. 2.Innovative and Spatially-Aware Architecture: While many existing MPI reconstruction frameworks—especially those based on standard CNNs or slice-wise diffusion—often treat 2D slices in isolation and neglect the critical inter-slice relationships, the authors propose a promising DiT-based architecture. By utilizing parallel PST and AST blocks to provide conditional guidance, this framework effectively captures axial dependencies, ensuring that the generated priors maintain both in-plane details and axial consistency. 3.Significant performance improvements and Clinical Potential: Under 4x sparse System Matrix conditions, Table 1 shows that BiSA-DiT achieves significant improvements in PSNR, SSIM, and nRMSE across 6 baselines. The low standard deviations observed across various cases demonstrate the model’s exceptional robustness and stability. Qualitative results in Figure 2 further show that the 3D MPI vascular images reconstructed by BiSA-DiT maintain high-fidelity details and greatcontinuity. These results suggest the framework can effectively address the bottleneck of prohibitively long SM calibration cycles, showing great potential for real-world clinical MPI applications 4.Strong Ablation Support for PST and AST Modules: The ablation study validates the essential roles of both PST and AST modules. Table 2 indicates that while each module individually surpasses the “Pure DiT” model, their integration yields the best results, successfully justifying the dual-branch spatial awareness design.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
The paper proposes BiSA-DiT, an iterative sparse-system-matrix MPI reconstruction framework that combines a diffusion transformer prior with an ADMM-based physics-driven data consistency module. The main architectural contribution is a bidirectional spatially-aware conditioning branch with planar spatial transformer blocks for intra-slice detail and axial spatial transformer blocks for inter-slice continuity.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.The paper addresses an important and practically relevant problem in MPI reconstruction, where reducing system-matrix calibration burden is highly desirable. 2.The visual results demonstrate clear improvements.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.The experimental setting is relatively limited in scale. The reconstructed volumes are only 32 x 32 x 24, which raises the question of whether a direct 3D diffusion model or a fully 3D learned prior could already be feasible at this resolution. Without such a baseline, it is difficult to determine whether the proposed PST/AST decomposition is necessary or whether the improvement mainly comes from using a stronger diffusion prior. 2.The comparison lacks several MPI-specific learning-based reconstruction baselines, including methods cited by the authors themselves, such as DEQ-MPI, U-N2C, and other related approaches for system-matrix denoising or sparse-SM reconstruction. 3.The robustness evaluation is limited, as the experiments only consider a single 4x sparse-SM setting. Sparse-SM reconstruction performance can strongly depend on the sparsity level, sampling pattern, noise level, and system-matrix mismatch. Additional experiments under different sparse-SM factors would be needed to demonstrate the stability of the method. 4.The method relies on supervised training with clean 3D ground-truth volumes derived from clinical CTA images. However, these GT volumes are not acquired from real MPI scans, but are synthetically generated through ROI extraction and projection from coronary CTA data. This raises concerns about the domain gap between CTA-derived vessel volumes and real MPI tracer distributions. 5.Relatedly, the paper does not sufficiently discuss out-of-distribution generalization. The learned diffusion prior is trained and evaluated on CTA-derived simulated MPI data with a fixed system matrix and a fixed sparse-SM setting. Its robustness to real MPI acquisitions, different anatomies, different tracers, scanner variations, noise levels, and system-matrix mismatch remains unclear.
Minor： 1.The paper refers to AST as both “Axial Spatial Transformer” and “Axial Structural Transformer.” The authors should use a consistent name throughout the manuscript.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I recommend a weak accept because the paper addresses an important and practical MPI reconstruction problem with clear visual improvements, although the current validation is limited by simulated CTA-derived data, small-volume experiments, incomplete MPI-specific baselines, and insufficient analysis of robustness and OOD generalization.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
The paper proposes BiSA-DiT, a diffusion-transformer-based framework for sparse-system-matrix 3D MPI reconstruction. The main idea is to combine a spatially-aware generative prior, built with planar and axial transformer blocks, with an ADMM-based data consistency module to improve both in-plane details and z-axis continuity.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.The paper addresses an important problem in 3D MPI reconstruction under sparse system matrices. 2.The method is well motivated and combines learned priors with physics-based consistency in a reasonable way. 3.The results are strong, with clear improvements over prior baselines. 4.The ablation study supports the benefit of the proposed PST and AST modules.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.The validation is restricted to simulated data derived from CTA images and OpenMPI system matrices. It remains unclear how well the method would generalize to real MPI acquisitions, measurement noise, calibration mismatch, or more challenging practical settings. 2.The paper evaluates only the 4× sparse SM setting, and it would be useful to see performance under varying sparsity levels or robustness to different reconstruction conditions. 3.Although the paper emphasizes learned spatial awareness, the physics-driven module still relies on handcrafted z-axis TV regularization. This somewhat weakens the claim that the method learns a fully expressive volumetric prior.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper tackles an important MPI reconstruction problem and proposes a technically reasonable framework that combines a spatially-aware diffusion-transformer prior with physics-driven iterative optimization. The empirical improvements are strong, and the ablation study supports the importance of modeling both planar detail and axial continuity. However, the methodological novelty is somewhat incremental, as the method mainly integrates existing ideas in diffusion reconstruction, DiT-style denoising, and ADMM-based consistency. In addition, the evaluation is limited to simulated data, so the practical robustness and real-world applicability remain to be further demonstrated.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
N/A
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
Despite some divergence in reviewers’ opinions, I recommend acceptance as the paper demonstrates good metrics and visual results alongside superior interpretability compared to other works. The technical design, specifically the integration of the ADMM solver and the parallel PST/AST modules, is fundamentally sound and well-justified.