Abstract

Positron Emission Tomography combined with Magnetic Resonance (PET-MR) imaging has emerged as a promising modality that offers both soft tissue and biochemical function information, while substantially reducing radiation exposure compared to PET-CT imaging. However, systematic clinical evaluations reveal notable discrepancies in standardized uptake value ratios between PET-MR and PET-CT scans, largely due to the inherent limitations of MR-based PET attenuation correction. To address this issue, we propose a unified uptake correction framework to harmonize PET-MR images with PET-CT scans across different tracers. This framework employs a three-stage training scheme. The first stage learns to represent CT features, aiming to capture condensed anatomical patterns associated with PET imaging. The second stage aligns MR features to the fixed CT features learned in the first stage, thereby enabling the transfer of anatomical prior knowledge from CT to MR features. The third stage integrates aligned MR features to guide PET-MR tracer uptake correction and uses a Multi-scale Pixel Routing module to mitigate interference among different tracers. We conduct comprehensive experiments on 70 patients with three distinct tracers to demonstrate the superiority of our framework over existing methods in PET-MR harmonization with PET-CT images. This work represents the first investigation and solution for multi-tracer quantification discrepancies between PET-MR and standard PET-CT, potentially advancing the clinical standardization of PET-MR imaging.Our code will be available at GitHub.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2476_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/ZAC0713/Multi-Tracer-PET-Uptake-Correction

Link to the Dataset(s)

N/A

BibTex

@InProceedings{ZhoAoc_MultiTracer_MICCAI2025,
        author = { Zhong, Aocheng and Huang, Haolin and Wang, Jing and Shen, Zhenrong and Song, Haiyu and Wu, Junlei and Zhu, Yuhua and Liu, Yang and Zuo, Chuantao and Wang, Qian},
        title = { { Multi-Tracer Uptake Correction for PET-MR via Aligned-Feature Guidance and Multi-scale Pixel-adaptive Routing } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15972},
        month = {September},
        page = {413 -- 422}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The paper presents a PET-MRI attentional correction model assisted by paired PET-CT. The method outperformed a few compared methods using local datasets of three tracers. The paper is well presented.

  • 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.

    The method outperformed a few compared methods using local datasets of three tracers.

  • 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.
    • The method requires paired PET-CT and PET-MRI which is a high requirement. It is unclear to what extent the proposed method can be used clinically. What’s the upper bound?
    • The baseline CBAM-block-based Unet already outperforms other compared methods. Are these compared methods were designed for the same purpose?
    • Statistical tests are required for all results (e.g. Wilcoxon signed ranked test), when claiming one method is better than another. Especially the number of test images are small.
    • While the target tracer is 3, why N=7 in the MsPRM is used? Does it work for more tracers?
  • Please rate the clarity and organization of this paper

    Satisfactory

  • 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • 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/2025/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 — could be accepted, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Not sure if the true SOTA methods were compared.

  • 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 #2

  • Please describe the contribution of the paper

    This paper proposes a solution to the problem of notable discrepancies in the standardized uptake value ratio between PET-MR and PET-CT. The cause of this problem is that the MR-based PET attenuation correction has inherent limitations. To address this issue, the authors of the paper propose a unified uptake correction framework to use PET-CT to harmonize PET-MR among different tracers. The authors put forward three steps: 1. Extract the structural features matching PET from CT; 2. Align the MR features with the CT features, thereby transferring the anatomical prior knowledge of CT to the MR features; 3. Use the MR features to guide the PET-MR uptake correction, and apply the multi-scale pixel routing module to transfer the interference among different tracers. In this paper, the effectiveness is verified and achieves results that are significantly better than SOTA algorithms.

  • 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.

    The unified uptake correction framework proposed in the paper is very novel. The paper makes full use of the advantage that CT images can better reflect anatomical features, and uses the network trained with CT images to assist in the feature extraction of MR images. Moreover, the paper collected data from 70 patients. These patients underwent examinations with three PET tracers, and both PET-CT and PET-MR scans were performed on them. We will greatly appreciate it if the authors can share these data.

    The experiments in the paper are relatively comprehensive, the writing is good, and the presentation is clear.

  • 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.

    The authors can further break down the problem and analyze for which types of data the proposed method has greater advantages, for which types of data the advantages are limited, and the reasons why.

    Please list some studies on the standardized uptake value ratio of PET - MR and PET - CT, as well as related correction methods, as the related work.

  • 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • 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/2025/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?

    The unified uptake correction framework proposed in the paper which utilize PET-CT image for PET-MR is very novel.

  • 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 a unified deep learning framework to correct multi-tracer quantification discrepancies between PET-MR and PET-CT, using a three-stage scheme: CT feature learning, MR-to-CT feature alignment, and a multi-scale pixel routing module for tracer-specific correction. This is the first work to address multi-tracer PET-MR harmonization with PET-CT.

  • 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. Novelty: First to address multi-tracer PET-MR/PET-CT harmonization.
    2. Technical Innovation: Effective CT-to-MR feature alignment and multi-scale routing.
    3. Comprehensive Evaluation: Tested on 70 patients with three tracers, showing consistent improvement over SOTA.
    4. Clinical Relevance: Potentially improves PET-MR standardization for clinical use.
  • 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. Dataset Size: 70 patients is modest; generalizability to larger or more diverse datasets is unclear.
    2. Computational Cost: Complexity and inference time not discussed.
  • 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • 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/2025/en/REVIEWER-GUIDELINES.html

    More discussion on computational efficiency and potential clinical workflow integration would be helpful. Consider adding qualitative failure cases.

  • 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?

    The paper addresses an important and underexplored clinical problem with a novel, well-validated approach. Despite some limitations in dataset size and clinical validation, the technical contribution and demonstrated effectiveness make it a strong candidate for acceptance.

  • 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

Thanks for the valuable comments. We appreciate the recognition of our work and have addressed each of the reviewers’ concerns as follows.

  1. Regarding the relevant work and comparison with SOTA methods (R1&R2&R3): As mentioned in the introduction, our study is the first to address the discrepancy in tracer uptake between PET-MR and PET-CT, a phenomenon often overlooked in clinical practice, and no previous studies have focused on this discrepancy or proposed a correction method. Therefore, our work is the first to highlight this issue and provide a feasible correction solution. We framed the tracer uptake correction problem as a low-level vision issue and compared it with SOTA image restoration (IR) methods widely used in natural images (SwinIR, Restormer). Additionally, we compared it with an advanced unified medical IR technique (AMIR) and a multi-center PET denoising study in the PET field (DRMC). By employing 5-fold cross-validation, our model achieved the best results in quantitative metrics (PSNR/SSIM) while also having the smallest computational complexity and the fastest inference speed. A more detailed analysis will be provided in the extended journal version.
  2. Regarding the results analysis and clinical applicability (R1&R2&R3): We sincerely appreciate the reviewers’ attention to the clinical applicability of our study. In our current results, we observed that the model performed significant corrections for the AV1 and TAU tracers, while the correction for the FDG tracer was relatively minor (Fig.3). This phenomenon highlights the tracer-specific discrepancies in PET-MR imaging (as mentioned in the introduction): compared to the PET-CT gold standard, PET-MR results for AV1 and TAU tracers exhibited statistically significant differences across multiple brain regions, while FDG tracer data showed detectable statistical differences only in a limited number of brain areas. Our method effectively reduces the statistical differences between PET-MR and PET-CT across various brain regions (Fig.1). Handling images of unseen tracers is likely the greatest challenge for the model. However, we have recently collected preliminary data on AV45, and we found that the model generalized well to unseen tracers, performing effective corrections. We will provide a more detailed analysis in the extended version.
  3. Regarding the limitation of the small paired dataset (R2&R3): We acknowledge this limitation. Since this is the first study addressing this issue, we prospectively collected 70 paired PET-CT and PET-MR datasets. Paired data is the most direct and accurate method for evaluating the correction performance of the model. However, the collection of such data is challenging. To mitigate this, we employed a 5-fold cross-validation to objectively assess the model’s correction effectiveness. Of course, we are actively collecting a larger set of data to more comprehensively validate the model’s performance and generalization capability.
  4. Regarding the number of experts in MsPRM (R2): We appreciate the reviewer’s insightful question. The number of experts in the MsPR module (N=7) is not directly related to the number of tracers. The initial design of MsPR takes into account that different tracers exhibit varying uptake patterns, which can lead to inter-tracer interference in the shared network parameters. Therefore, the primary purpose of MsPR is to eliminate this interference. The MsPR module assigns each pixel from the input PET-MR image to the most suitable top-K multi-scale expert networks based on routing instructions. The reason we chose N=7 is that, unlike traditional multi-scale designs that rely on repeated downsampling, we found that combining both upsampling and downsampling better captures fine-grained features. Consequently, we performed three rounds of upsampling and downsampling while preserving the original scale. Due to page constraints, we will provide a more detailed comparative analysis in the extended version.




Meta-Review

Meta-review #1

  • Your recommendation

    Provisional Accept

  • If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.

    N/A



back to top