List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Haoran Dou, Ning Bi, Luyi Han, Yuhao Huang, Ritse Mann, Xin Yang, Dong Ni, Nishant Ravikumar, Alejandro F. Frangi, Yunzhi Huang
Abstract
Deep learning-based deformable registration methods have been widely investigated in diverse medical applications.
Learning-based deformable registration relies on weighted objective functions trading off registration accuracy and smoothness of the deformation field. Therefore, they inevitably require tuning the hyperparameter for optimal registration performance. Tuning the hyperparameters is highly computationally expensive and introduces undesired dependencies on domain knowledge. In this study, we construct a registration model based on the gradient surgery mechanism, named GSMorph, to achieve a hyperparameter-free balance on multiple losses. In GSMorph, we reformulate the optimization procedure by projecting the gradient of similarity loss orthogonally to the plane associated with the smoothness constraint, rather than additionally introducing a hyperparameter to balance these two competing terms. Furthermore, our method is model-agnostic and can be merged into any deep registration network without introducing extra parameters or slowing down inference. In this study, We compared our method with state-of-the-art (SOTA) deformable registration approaches over two publicly available cardiac MRI datasets. GSMorph proves superior to five SOTA learning-based registration models and two conventional registration techniques, SyN and Demons, on both registration accuracy and smoothness.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43999-5_58
SharedIt: https://rdcu.be/dnwxb
Link to the code repository
https://github.com/wulalago/GSMorph
Link to the dataset(s)
https://www.creatis.insa-lyon.fr/Challenge/acdc/databases.html
Reviews
Review #2
- Please describe the contribution of the paper
This paper construct a registration model based on the gradient surgery mechanism, named GSMorph, to achieve a hyperparameter-free balance on multiple losses. Instead of introducing a additional hyperparameter to balance two competing terms, GSMorph project the gradient of similarity loss orthogonally to the plane associated with the smoothness constraint. GSMorph proves superior to compared methods on both registration accuracy and smoothness.
- 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 article is well organized and easy to follow, the writing is good.
- The GSMorph presents a gradient-surgery-based registration framework for medica images. This study employ gradient surgery to refine the optimization procedure in learning the deformation fields which is innovative.
- The model outperformed the DLR models trained with other GS-based methods. And, its results are comparable to those of other method which show the effectiveness of the method.
- The method can be easily integrated into any DLR network without extra parameters or losing inference speed.
- 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.
- Although the method is better than GS-based models, its performance do not show its superior to the model optimized by conventional method.
- Please rate the clarity and organization of this paper
Very 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 of the paper will be good, because authors claim the the source code will be available online soon.
- 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 registration network is based on 2D images or 3D images？If it is based on 2D, what is your estimate of the performance when it is extended to 3D?
- I prefer to see the error maps between the target, the moving and the warped images in Fig. 3.
- Which similarity loss is actually used? I think more experiments about different loss functions would be interesting in the future works.
- 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
7
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The idea is simple and effective. The results is comparable.
- 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
Review #3
- Please describe the contribution of the paper
Tuning the hyperparameters to balance the accuracy and smoothness for learning based deformable registration requires highly computational cost. This paper proposed to use gradient surgery mechanism to achieve a hyperparameter-free balance on multiple losses. Specifically, they project the gradient of similarity loss orthogonally to the plane associated with the smoothness constraint. Compared to other methods on the cardiac MRI dataset, the proposed method achieves superior performance.
- 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.
They try to tune the hyperparameter using gradient surgery mechanism. Specifically, they project the gradient of similarity loss orthogonally to the plane associated with the smoothness constraint. Compared to other methods on the cardiac MRI dataset, the proposed method achieves superior performance.
- 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 performance may be better on brain MR?
- Please rate the clarity and organization of this paper
Very 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
ok
- 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
N/A
- 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?
The idea is quite good, and results are ok.
- 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 #1
- Please describe the contribution of the paper
The proposed GSMorph addressed the balance of multiple terms to accelerate development and deployment of DLR models. They provide a geometric view to depict the gradient changes for θ based on the gradient surgery technique, and avoid hyperparameter tuning while training the DLR. Gradient surgery (GS) projects conflicting gradients of different losses during the optimization process of the model to mitigate gradient interference. This work used the GS strategy [15, 20] for the image registration and has been evaluated on ACDC and M&M datasets.
- 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 proposed GSMorph addressed the balance of multiple terms to accelerate development and deployment of DLR models. They provide a geometric view to depict the gradient changes for θ based on the gradient surgery technique, and avoid hyperparameter tuning while training the DLR. Gradient surgery (GS) projects conflicting gradients of different losses during the optimization process of the model to mitigate gradient interference. This work used the GS strategy [15, 20] for the image registration and has been evaluated on ACDC and M&M datasets.
- 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.
There lack of discussions on how the proposed GSMorph to ensure the regularization be satisfied in the predicted displacement fields. The g_reg does not contribute to the update of the parameters θ in the proposed algorithm. It would be helpful to discuss the performance gains over the compared methods, especially with respect to the NJD.
- 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 of the paper is OK.
- 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 author claimed that “The conflicting relationship between two controversial constraints can be geometrically projected as orthogonal vectors.” As shown Fig. 2, the projection is used to remove the conflicting components from g_sim. Since the g_reg does not contribute to the update of Parameters θ, it is not clear how to ensure the regularization is satisfied in the predicted displacement fields.
-
As shown in Table 1， the VM with a predefined \lambda showed the prevailing performance on ACDC. Does it mean that the proper weighted combination of the similarity and the regularization terms is preferable to optimize the registration network?
-
Also in Table 1, there seems to be a performance gap between dataset ACDC and MM. The proposed GSMorph outperformed the VM on the DSC, and even have smaller NJD compared with the VM-s,m. Considering the gradient orthogonal to g_reg in the proposed method, I suggest a discussion on the performance gains, especially regarding the NJD.
-
Table 2 reports the inference speed of the compared method. The proposed method has similar complexity to the compared deep registration models. I am not sure whether the speeds of all compared methods are in the same units. It seems that the conventional registration method, such as Demons is also efficient for online registration.
-
- 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?
The proposed GSMorph addressed the balance of multiple terms to accelerate development and deployment of DLR models. This work used the GS strategy [15, 20] for the image registration and has been evaluated on ACDC and M&M datasets. There lack of discussions on how the proposed GSMorph to ensure the regularization be satisfied in the predicted displacement fields. It would be helpful to discuss the performance gains over the compared methods, especially with respect to the NJD.
- 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.
All reviewers acknowledge the strength of the method, which employs the gradient surgery mechanism to avoid hyper-parameter tuning between multiple loss terms in deep learning registration models. The idea is simple yet effective, which could perhaps be adopted in other medical imaging applications beyond registration. Overall, the paper is well organised and easy to follow.
Author Feedback
We thank the reviewers for the constructive comments. We will revise them in our journal version paper.