Paper Info Reviews Meta-review Author Feedback Post-Rebuttal Meta-reviews

Authors

Ruisheng Su, Matthijs van der Sluijs, Sandra Cornelissen, Wim van Zwam, Aad van der Lugt, Wiro Niessen, Danny Ruijters, Theo van Walsum, Adrian Dalca

Abstract

Cerebral X-ray digital subtraction angiography (DSA) is the standard imaging technique for visualizing blood flow and guiding endovascular treatments. The quality of DSA is often negatively impacted by body motion during acquisition, leading to decreased diagnostic value. Traditional methods address motion correction based on non-rigid registration and employ sparse key points and non-rigidity penalties to limit vessel distortion, which is time-consuming. Recent methods alleviate subtraction artifacts by predicting the subtracted frame from the corresponding unsubtracted frame, but do not explicitly compensate for motion-induced misalignment between frames. This hinders the serial evaluation of blood flow, and often causes undesired vasculature and contrast flow alterations, leading to impeded usability in clinical practice. To address these limitations, we present AngioMoCo, a learning-based framework that generates motion-compensated DSA sequences from X-ray angiography. AngioMoCo integrates contrast extraction and motion correction, enabling differentiation between patient motion and intensity changes caused by contrast flow. This strategy improves registration quality while being orders of magnitude faster than iterative elastix-based methods. We demonstrate AngioMoCo on a large national multi-center dataset (MR CLEAN Registry) of clinically acquired angiographic images through comprehensive qualitative and quantitative analyses. AngioMoCo produces high-quality motion-compensated DSA, removing while preserving contrast flow. Code is publicly available at https://github.com/RuishengSu/AngioMoCo.

Link to paper

DOI: https://doi.org/10.1007/978-3-031-43990-2_72

SharedIt: https://rdcu.be/dnwMw

Link to the code repository

https://github.com/RuishengSu/AngioMoCo

Link to the dataset(s)

N/A


Reviews

Review #1

  • Please describe the contribution of the paper

    This article predicts blood vessels through the first network and removes them from DSA images. Then, a second network is used to predict the deformation field between the image after removing blood vessels and the image without DSA. Explicitly constructed the process of removing blood vessels, improving the effectiveness of the method. I think this is the main innovation of this article.

  • Please list the main strengths of the paper; you should write about 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.

    Explicitly constructed the process of removing blood vessels, improving the effectiveness of the method.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    The report on the results in the article is not clear enough to intuitively observe the advantages of the method. This article should include detailed numerical values of the statistical results in the table for greater clarity. If the result table is not provided in subsequent versions, I will reject this article.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance

    “We will make the code publicly available.”in this paper.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html

    The report on the results in the article is not clear enough to intuitively observe the advantages of the method. This article should include detailed numerical values of the statistical results in the table for greater clarity. If the result table is not provided in subsequent versions, I will reject this article.

  • Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making

    5

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

    Explicitly constructed the process of removing blood vessels, improving the effectiveness of the method.

  • Reviewer confidence

    Somewhat confident

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #2

  • Please describe the contribution of the paper

    This article propose to improve the image processing for digital subtraction angiography (DSA). DSA is a very ancient technique which aims to increase the benefit obtained by the injection of a contrast agent by subtracting from the current image, an image taken before the injection of contrast. In the absence of motion, the resulting subtracted image offers an excellent visualization of the vessels. In presence of motion, subtraction artifacts degrades the visibility of the structures of interest. The authors proposes an approach based on deep-learning. A first module generates an image without contrast from an image post-injection. An other module perform a motion correction between this image and the image before the injection. The set is then logically recombined to obtain a subtracted image with reduced motion artifacts.

  • Please list the main strengths of the paper; you should write about 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 authors revisited a well-known problem applying new machine learning strategy while keeping the classical philosophy : contrast extraction & motion estimation being separated which makes sense. They have also worked with a pretty large database of clinical data.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
    • The authors mention several times a possible issue with different methods including their own method. It is named “vessel distortion”. It is neither defined which makes the understanding of the article difficult;
    • From a methodology standpoint, the different images are normalized to [0 1] with a min/max approach. It is a rather poor approach that entirely neglects that the physics of the image formation. In a DSA series, the consecutive images are obtained with stable x-ray technics and so the intensity are coherent along the series. The normalization process retained by the authors disrupts this coherency if the min/max changes from one image to the other. This is very likely to happen due to the noise.
    • The subtraction is done with a simple subtraction. The authors does not mention any other method. It is well known that given the law of Beer-Lambert applicable to x-ray image, the application of a log transformation to the image is very useful and allows to have consistent contrast in the images
    • The article title is misleading versus the experimental work conducted. The authors have work on cerebral angiography. It is a subset of DSA field. DSA is also applied to lower and upper limbs and also to abdomen. In this later case, the presence of intestinal motion makes especially challenging the problem of motion artifacts in DSA. To avoid any confusion, I strongly recommend that the article title and the abstract clarifies the content. Indeed, this specificity is only mentioned page 5.
    • There is an assertion that the new proposed method is much faster than more traditional methods. This assertion is grounded on an unfair comparison. One method is implemented on GPU and the other is referenced to be run on a CPU. It does not tell that the traditional method cannot be run on GPU. So this points shall be clarified by providing for example a comparison of the times taken on the same hardware. I would recommend CPU since this is the classical support hardware for this type of algorithm in a product implementation. For multiple reasons including costs, GPU are not so widely present in the product involved with DSA.
    • Quantitative experimental results are obtained by measuring the Mean Subtracted Intensity at different places in the images: in and out of the vessels. The authors mentions that the vessels have been manually delineated. The explanation is a bit short and there is no illustration to get a feeling of the performed delineation. Is it the full vessel set of an image including the largest and the smallest vessel? Is it only a central line or the entirety of the vessels? We are again back to the already mentioned concept of vessel distortion which has not been explained. We ignore if this quantification allows to capture such issue.
  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance

    The reproducibility is possible knowing that the authors mention that they will make the code available.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html

    As a general comment and improvement directions, I would recommend that you spend sometimes to take into account the physics of the x-ray: noise distribution, scatter, contrast as a function of kV, vessel diameter, etc. Considerations shall also be given to the processing of the images done by the image manufacturer before being exported as Dicom images. All these elements may have an effect on the possible processing.

  • Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making

    5

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

    I have found the investigation interesting and focus on an important problem. I regret that some terms like “vessel distortion” were not explained. I also regret that the authors treated the problem as an image processing problem and did not investigate more in depth the physics of the imaging as well the effect of the motion on the x-ray image. Other elements such as medical needs were a little bit overlooked and summarized by a measure like “mean square intensity” . One may think that the relative importance of large and small vessel is not the same at all. Subtraction artifacts may be simply annoying or a true problem for image interpretation. They shall certainly being quantified differently. I have not seen these points being addressed.

  • Reviewer confidence

    Very confident

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #3

  • Please describe the contribution of the paper

    The paper presents a learning-based approach for motion correction in DSA. The authors claim that it can significantly improve the quality and diagnostic value of the imaging results.

  • Please list the main strengths of the paper; you should write about 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.
    • Interesting application
    • Once trained, the framework is quick to execute compared to classic non-rigid registration methods
    • Good comparison against existing methods
    • good methodology
  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
    • The different models are not trained end-to-end. It would have been nice to see this. If this is not possible due the dataset split (motionless/motion), then it should not be presented as a single framework, but rather two separate models trained on two separate datasets.
    • It is unclear how the separation between motionless and motion subsets is operated. “visual assessment” is not enough. More quantitative and objective criteria should be used.
    • The contribution of the regularization term is not well discussed.
  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance

    Good reproducibility of the neural networks method. The author state that the data will be released.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html

    A more quantitative criterion should be adopted for separating the motionless/motion datasets. Presenting the method as a single, unified workflow, may be confusing, since the two networks are trained separately. Additionally, the overview figure (2) could be modified: the output of f_{\theta_f} could be represented differently (single output and then to m_t).

  • Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making

    5

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

    good overall methodology. good comparison against other methods. poor dataset definition and workflow representation.

  • Reviewer confidence

    Confident but not absolutely certain

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A




Primary Meta-Review

  • Please provide your assessment of this work, taking into account all reviews. Summarize the key strengths and weaknesses of the paper and justify your recommendation. 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.

    The manuscript describes a method for motion-related artifact reduction in cerebreal digital subtraction angiography.

    The paper has received 3 weak accept recommendations, so there is conensus among the reviewers.

    While all reviewers appreciated the approach to problem modeling in that it followed the tried-and-trusted DSA workflow, that the method was evaluated on large clinical data, and that code will be made available (there seems to be perceived discrepancy, as one reviewer mentions that data will also be released - clarification needed).

    The reviewers also mentioned several concerns, some of them quite detailed, that should be addressed as possible in the next revision. The most important ones include:

    • Contextualization around not further considering physical undertsanding of image formation;
    • Some overstated claims (for example around the inference speed due to inappropriate comparison),
    • Lack of detail in results presentation (R1 has a strong concern around this).

    Several of the other issues raised during review should also be addressed, but those should be straightforward.




Author Feedback

We thank the reviewers and the area chair for the positive and productive feedback. We address several raised questions below.

  • Availability of code and data [R3] The code will be made available. Due to constraints of clinical data, we can unfortunately not release the clinical imaging dataset.
  • Physics of image formation [R2] We agree that we should improve the normalization description. We performed min-max normalization on the series level, rather than per frame. We observed that while the maximum intensity slightly varies across different series, it is always close to the 2^BitsStored in the dicom. Therefore, instead of using the maximum intensity of the series, we used the “BitsStored” field in each dicom file as the maximum. The coherency of the intensity along the series is thus not disrupted.
  • Detailed result presentation [R1] Detailed numerical experiment results are presented In Table 1 of the supplement. We will include additional statistical details (present in the article) in the table for greater clarity.
  • Inference speed [R2] As suggested by R2, we will clarify the inference speed and computational resources used in the final version.
  • Vessel delineation [R2] We will include example figures in the supplement to illustrate the manual delineation of vessels.



back to top