Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Image-based biomarkers provide non-invasive regional assessment of structural-functional abnormalities in Chronic Obstructive Pulmonary Disease (COPD). For example, quantitative computed tomography (QCT) identifies emphysema and small airway disease, while functional MRI measures lung ventilation and perfusion. In recent years, machine learning techniques have been introduced to predict quantitative indices from alternative imaging modalities, with the aim to reduce scanning time, radiation dose and/or costs in the clinical setting. However, most of those works focused on lung ventilation, while robust quantification of regional lung perfusion of dynamic contrast-enhanced (DCE) MRI remains a challenging task. In addition, previous studies focused only on learning from a single imaging modality. In this study, we explore a deep learning-based model to predict conventionally CT-based biomarkers from multi-sequence structural-functional MR images. Our proposed model achieves very strong correlations in predicting PRMemphysema(Pearson correlation coefficient r = 0.91, p < 0.001 at patient level and r = 0.87, p < 0.001 at lung lobe level), and moderate to strong correlations in predicting %PRMnormal (r = 0.60, p < 0.001 at patient level and r = 0.58, p < 0.001 at lung lobe level) in unseen COPD patients.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/4198_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/YilingMed/MR2PRM4COPD

Link to the Dataset(s)

N/A

BibTex

@InProceedings{XuYil_ALearning_MICCAI2025,
        author = { Xu, Yiling and Triphan, Simon M. F. and Grolig, Julian and Zhang, Hanyi and Biederer, Jürgen and Galbán, Craig J. and Kauczor, Hans-Ulrich and Wielpütz, Mark O. and Weinheimer, Oliver},
        title = { { A Learning Framework for Predicting CT-based PRM Biomarker from MRI Sequences in COPD } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15967},
        month = {September},

}

Reviews

Review #1

  • Please describe the contribution of the paper

    The authors proposed a novel method to predict the parametric response map (PRM) classifications of the parenchyma (emphysema, functional small-airways disease, and normal parenchyma) on MRI sequences (structural and functional) for COPD. Additionally, the authors proposed to use an nnUnet trained from scratch to identify the lung lobes for posterior analysis of the PRM classification on lobes.

  • 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 fuses functional and structural MRI information to improve the prediction of the PRM classification.
    • The proposed method is not novel itself, but its application is.
    • The results demonstrate a significant improvement over existing MRI-based methods for PRM classification.
  • 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.
    • Limited novelty and contribution.
    • A small dataset was used for testing and measuring the model’s performance (N=30).
    • Registration plays a critical role in the performance of the proposed method; however, no details are provided. Could the authors specify the type of registration used to map the Post-Contrast VIBE, IRFmax, and PBF images to the VIBE reference image? An affine registration would typically be expected over a free-form deformation—please clarify which approach was used and explain the rationale behind this choice. The same clarification is requested for the registration process used to align the CT-based lobe segmentation with the MRI space.
    • The PRM is computed on CT scans and subsequently co-registered to the MRI space, where it is used as the ground truth (GT) in the loss function. This metric is susceptible to partial volume effects due to the lower resolution of the MRI, as the authors have noted. However, it is not immediately clear how the three proposed loss functions help to minimize this effect. One could speculate that label smoothing may mitigate registration errors in the PRM measurements, but it is unlikely to address partial volume effects since smoothing is applied after registration, according to the loss formulation. Could the authors clarify the rationale behind their statement on how the loss functions help address the partial volume effect? Additionally, could they explain in more detail what is being computed in the referenced equation and how the standard and blurred cross-entropy (CE) terms are aggregated across the image? For instance, is the loss computed by averaging the CE at each pixel?
    • Table 1 reports the correlation values for the QDP method [13]. However, these values differ from those originally reported by the authors of [13] (see Table 4 in [13], where r_emph ranges from 0.75 to 0.70 and r_fSAD from 0.34 to 0.37 across the methods evaluated). Could the authors clarify this discrepancy?
  • 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 submission does not provide sufficient information for reproducibility.

  • 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

    – The paper is relatively concise and easy to follow. – The authors do not provide enough information to reproduce the models proposed or the results obtained in this work. The model architecture is partially described; for example, how is the Feature attention layer implemented?

    • Describe acronyms before using them, for example, PRM.
    • I would suggest for future works the use of techniques such as Bootstrap to compute p-values and CI95%, especially for limited testing datasets.
    • The text in Fig.1 is too small.
    • Bland-Altman plots: It is useful to plot the +-1.95sigma values.
    • Typos: o Extra “)” on page 4. o supra-indices are missing on L_PRM loss, change class -> c
  • 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 — could be rejected, dependent on rebuttal

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

    The method proposed is partially described and it shows limited novelty. The validation was performed on a small number of subjects for an application study.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    All the concerns were addressed but there is still a limited novelty and contribution



Review #2

  • Please describe the contribution of the paper

    This study explored a deep learning-based model to predict conventionally CT-based biomarkers from multi-sequence structural-functional MR images. The proposed model achieves very strong correlations in predicting %PRMemphysema , and moderate to strong correlations in predicting %PRMnormal in unseen COPD patients.

  • 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 major strength of the paper is the demonstration of clinical feasibility with good statistical analysis - first time that using multi-sequence approach to predict CT-based biomarkers. Good results demo and interpretation.

  • 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. Further explanation needed for usage of CT images, any chance the author could build a pure end-to-end MRI framework for such prediction.
    2. Major 2 approaches - nnUNET and Synthesis-based imaging-differentiation representation learning for multi-sequence 3D/4D MRI are not original. Also Github resource codes are not provided.
  • 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 submission does not provide sufficient information for reproducibility.

  • 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?

    Overall the major strength of the paper is its clinical feasibility, I would suggest to strengthen clinical significance during rebuttal.

  • 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 #3

  • Please describe the contribution of the paper

    The paper introduces a deep learning-based framework designed to predict CT-based Parametric Response Mapping (PRM) biomarkers from multi-sequence structural-functional MRI in COPD patients. This framework combines lung lobe segmentation using nnUnet and a modified encoder-decoder-based image model to perform voxel-wise PRM classifications. By leveraging complementary information from multiple MRI sequences, the model achieves strong correlations with CT-derived PRM emphysema and moderate correlations with PRM normal

  • 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. By reducing dependence on CT scans, the study targets a specific clinical challenge and provide a feasible solution.

    2. The paper provide a relatiive clear description of the total pipeline. The whole module design is reasonable.

    3. The results can be promising, while the visualized resulrs can also provide supports.

  • 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 study is conducted on a relatively small dataset of 104 COPD patients, which may limit the generalizability of the results and the model’s robustness when applied to larger, more diverse populations. Different from segmentation, the regression task can be relatively less supervised; the small size of datasets may lead to less robust results.
    2. The proposed framework aims to use multi-sequence MRI data when the CT data is nnot avaliable; however, the multi-sequence MRI data can rather be less avalible in clinical practices. This could impair the clinical application of this method.
    3. The framework requires advanced image registration. I concern that when the registration method results to a inferior performance, how does the model performance be impaied?
    4. Although the clinical target is clear, the overall technique contribution and experiments can be limited.
  • 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 submission does not provide sufficient information for reproducibility.

  • 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?

    The manuscript is overall clear and I think it could be published; however, there are still many shortcomings and room for improvement, as I mentioned above.

  • Reviewer confidence

    Somewhat confident (2)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    The response can partially address my concerns, while the contribution of this paper is overall meaningful.



Author Feedback

We thank all reviewers for their constructive feedback, especially for recognizing our paper’s contributions to clinical applications and promising results. Our responses to major concerns are as follows: 1) Clinical Feasibility (R3, R4): The major motivation of the work is to limit cumulative ionizing radiation in COPD monitoring, given its proven association with elevated cancer risks. While we acknowledge R4’s comment that multi-sequence MRI is not yet a clinical standard, our pipeline offers a radiation-free alternative to predict established PRM biomarkers. Moreover, we demonstrated the model’s feasibility with missing sequences. We would also like to point out that paired same-day CT and MRI scans in COPD are rare. Our evaluation data size is comparable to related work such as n=47 in [2] and n=44 in (Liu et al. IJCARS 2025). For this study, only a subset of 104 individuals was examined, all using the same MRI scanner model, and only baseline data from the entire study were used. We plan to expand our dataset. 2) Usage of CT images (R3): We clarify that the ground truth of lobe segmentation and voxel-wise PRM classification were generated from CT images, and they were used during the training phase (Section 2.2, Fig. 1). The final model predicts exclusively from MRI to replace CT images, ensuring a radiation-free end-to-end approach. 3) Image registration (R1, R4): We acknowledge its importance and now provide further implementation details: TWIST MRI (used to compute PBF and IRFmax) and post-contrast VIBE images were registered to morphological VIBE MRI using affine and deformable registration, with aim of correcting minor positioning or breathing state shifts. CT images, as acquired at higher resolution and potentially different field-of-view, were first resampled to match the spacing of VIBE MRI, and subsequently aligned using ANTspy SyN framework which consists of rigid, affine and deformable registration. The computed spatial transformation was hence applied to lobe segmentation and PRM maps as well. We will ensure those details to be updated in the manuscript. We further emphasize that CT images were only used for evaluation in the test phase. Therefore, the most challenging registration step - between CT and MRI - is only necessary for voxel-wise evaluation. Specifically, the results of patient-level and lobe-level comparisons are independent of registration performance. 4) Loss function (R1): We clarify that the smoothing operation does not directly reduce partial volume effects (PVE), but is intended to stabilize the training process against its influence by taking neighboring voxels into consideration. The rationale is to encourage the model to focus on lung functional units rather than voxel-wise labels. Leaving out this loss leads to a noticeable decline in model performance, for example, the correlation with PRM_normal decreases by >0.1. For implementation, we applied Gaussian blurring to all three channels of the one-hot-encoded ground truth and computed the cross-entropy loss on the smoothed labels. R1 also mentioned the possibility of a smoothing operation before registration, which is an interesting idea that we would like to test out in future work. 4) Comparison with existing biomarkers: R1 sharply noted the difference between correlation coefficient r between QDP and PRM in Table 1 and reported in [13]. This can be explained by the different subsets of subjects used: Results in [13] was computed on 76 subjects from a single center, while our results computed on the randomly split test set (n=30) from 3 imaging centers. We computed 95%CI interval of r_emph is [0.32, 0.80] and r_fSAD is [-0.51 0.19] for QDP, compared to [0.81, 0.96] and [-0.22, 0.49] respectively for our proposed model. We will carefully polish the manuscript and release the code to ensure clarity and reproducibility.



Meta-Review

Meta-review #1

  • Your recommendation

    Invite for Rebuttal

  • 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”.

    This paper on predicting normally CT-based biomarkers such as parametric response map classifications of the lung parenchyma on multi-modal MRI sequences in patients with COPD was reviewer by three reviewers leading to mixed reviews. A rebuttal seems hence an appropriate next step.

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    N/A



Meta-review #2

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Reject

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    Although reviewers give this paper positive scores (1 reviewer has low confidence), they raise several major concerns in the reviews. After reading the paper, reviews and rebuttal, I found several major flaw of this work. 1) The overall novelty and contribution is limited. 2) The testing dataset is small. 3) Both quantitative results and qualitative results are weak. E.g., in Fig.2, predicted PRM have large difference with GT PRM, which indicates the unreliable measurement of the method. Hence, I recommend rejection.



Meta-review #3

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    Reviewers agree on the merits of the paper outweighing the limitations. Several weaknesses were convincingly addressed in the rebuttal. Whereas the methodological contribution is small, the relevance of the method seems good. If accepted, the nature of the small training dataset should be highlighted as a limitation in the final manuscript, as it was mentioned by several reviewers.



back to top