List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Jiangchang Xu, Yining Wei, Huifang Zhou, Yinwei Li, Xiaojun Chen
Abstract
Orbital blow-out fracture (OBF) is a complex disease that can cause severe dam-age to the orbital wall. The ultimate means of treating this disease is orbital wall repair surgery, where automatic reconstruction of the orbital wall is a crucial step. However, accurately reconstructing the orbital wall is a great challenge due to the collapse, damage, fracture, and deviation in OBF. Manual or semi-automatic re-construction methods used in clinics also suffer from poor accuracy and low effi-ciency. Therefore, we propose a symmetric prior anatomical knowledge (SPAK)-guided generative adversarial network (GAN) for automatic reconstruction of the orbital wall in OBF. Above all, a spatial transformation-based SPAK generation method is proposed to generate prior anatomy that guides the reconstruction of the fractured orbital wall. Next, the generated SPAK is introduced into the GAN network, to guide the network towards automatic reconstruction of the fractured orbital wall. Additionally, a multi-function combination supervision strategy is proposed to further improve the network reconstruction performance. Our eval-uation on the test set showed that the proposed network achieved a Dice similari-ty coefficient (DSC) of 92.35±2.13% and a 95% Hausdorff distance of 0.59±0.23mm, which is significantly better than other networks. The proposed network is the first method to implement the automatic reconstruction of OBF, ef-fectively improving the reconstruction accuracy and efficiency of the fractured orbital wall. In the future, it has a promising application prospect in the surgical planning of OBF.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43996-4_44
SharedIt: https://rdcu.be/dnwPp
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes a symmetric prior anatomical knowledge (SPAK)-guided generative adversarial network (GAN) for more accurate reconstruction of the orbital wall in orbital blow-out fracture (OBF).
- 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 symmetric prior anatomical knowledge guidance could be automatically generated, then incorporated into GAN-based orbital wall reconstruction.
- 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 should elaborate on how the spatial transformation-based symmetric prior knowledge module ensures optimal symmetric performance. The authors could add some quantitative experiment results.
- As U-Net and V-Net may now be considered a little bit outdated, the authors could use some modern U-Net-based benchmarks for fair comparison.
- 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 paper’s main concept is clear and I have confidence in its ability to be reproduced.
- 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
Some comments have been listed in the above sections.
- 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
4
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The method’s unclear descriptions and the lack of some comparison experiments may not persuade me to consider accepting it.
- 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 #4
- Please describe the contribution of the paper
The paper describes a method for estimating orbital wall geometry in subjects with orbital blow out fractures. The method uses a high resolution computed tomography volume of the skull, consistent with the current standard of care for surgical planning. The authors propose a novel technique that uses “symmetric prior anatomical knowledge (SPAK)” by concatenating the uninjured image and orbital wall segmentation with the fracture orbit image and segmentation in a GAN framework. The approach and problem are both novel. The authors compare the suggested network to other established methods for generating fixes to defects in bone and show their method has 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.
- Well written and clear text.
- Abstract is succinct and has relevant details
- Novel application with a good justification
- A novel approach that aims at using the inherent symmetry of the contralateral side to improve performance of restored fractured geometry
- Ablation studies are through and show the SPAK
- 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 main weakness of the work is that aspects of the work could be better explained leading to some confusion o Is the entire network trained end to end o Patches within the cropped orbital regions were used because of computational requirements. Or are the patches referred to the orbital regions cropped out of the skull ct volumes?
- Please rate the clarity and organization of this paper
Excellent
- 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 things about reproducibility o well described methods o clear pre and post processing described o
- limiting the ability to reproduce o a closed dataset is used o The method for manual annotation of the images could be better described. Little is said about how the segmentations were done than an experienced clinician did them. It is unlikely that another clinician would exactly segment and reconstruct as done in this investigation
- 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
- Consider reworking the first sentence. Is it true that transportation popularity is growing? If so please reference.
- BOF should be changed to OBF
- It would be insightful to have a comment in the discussion about resolution. The orbital walls are thin. How thick are the segmentations in this investigation? How does this affect the results? How well resolved is the orbital wall after resampling for the network? What is the nominal voxel size after resampling?
- 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
8
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
novel application and a novel approach with useful ablation and comparison studies
- 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
7
- [Post rebuttal] Please justify your decision
The comparisons to other approaches could have been more thorough and the establishment of how the SPAK works could have been more clear in the explanation and with quantitative experiments.
Review #5
- Please describe the contribution of the paper
The authors presented the application of a deep learning model to automatically reconstruct the orbital wall when one of the orbital walls is fractured. To achieve that, they first developed a method to generate prior anatomical information, i.e., the symmetrical shape of both orbital walls, based on the shape of the non-fractured orbital wall. Then, they applied GAN to generate the normal shape of the fractured orbital wall based on the obtained prior information. The ablation study shows that the anatomical prior information is useful to improve the performance of GAN.
- 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.
- Novel Application. The application of the GAN model to reconstruct fractured orbital wall is novel, especially with the incorporation of prior anatomical information into the GAN model.
- Clinical Feasibility. The proposed work will help surgeons to easily understand the desired shape of the fractured orbital wall.
- Clear Experimental Setup. Authors clearly described the how the dataset was collected, including the quality of the CT scans and implementation details.
- 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.
- Limited Clinical Application. This work is not applicable when both orbital are fractured.
- Limited Technical Contribution. Though the application of the anatomical-prior based GAN model to reconstruct orbital wall is novel, this work uses existing framework (3D V-Net and GN-DN) with a few additional mechanical steps.
- Vague Model Evaluation. It is unclear how the authors used U-Net, V-Net, Attention U-Net, and Attention V-Net for surface reconstruction.
- Please rate the clarity and organization of this paper
Satisfactory
- 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 paper contains small amount of implementation details. However, the authors will provide code, which may improve the reproducibility of the 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
MAJOR
- “including (a) medial wall fracture, (b) floor wall fracture, (c) fractures in both the medial and floor walls, (d) roof wall fracture, (e) and other types.” How is the distribution of the dataset on these types of OBF?
- In Section 2.1, please provide a citation on studies that found the orbital walls are (somewhat) symmetrical to validate the assumption for SPAK.
- After reconstruction is there any change on the non-fractured region?
- Missing explanation on how the ICP in SPAK is performed. The primary concern is that ICP is sensitive to initial registration between the two 3D objects. Also, it is unclear how the deformation field is obtained.
- Please provide details on the concatetation in GAN.
- Please elaborate on how U-Net, V-Net, Attention U-Net, and Attention V-net are used for surface reconstruction.
- Table 1 shows insignificant improvement when DN is added to GN+SPAK. What’s the possible reason for this?
- How will the different cropping affect the performance of the model? It is important because the cropping is performed manually.
MINOR
- Suggestions.
- Fig. 1 is difficult to be understood. Please consider a different way of annotating the region of interest. It may be helpful to show a normal and fractured images for each type of OBF.
- IMHO, 95HD is representative of surface distance, thus ASD can be removed.
- 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
4
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Limited Technical Contribution. Limited Clinical Application. Vague Model Evaluation.
- 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
5
- [Post rebuttal] Please justify your decision
Authors provided reasonable justification on the Method innovation and clinical application of the paper (except the 3D slicer integration aspect) and dataset details. Note that providing a citation to “ as unilateral orbital fractures constitute over 95% of all cases” is required.
Reviewers agreed on the usefulness of SPAK, but authors does not provide the useful addtional information (e.g., ICP, global registration, deformation field) on its technical details during the rebuttals.
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 paper proposes a SPAK-guided GAN model for automatic reconstruction of the orbital wall in OBF surgery. However, the quality of the figure is so poor, for example Fig. 1, and 2. In Table 2, more transformer-based network should be included in the comparative experiments.
Author Feedback
Thanks to the meta-review and all reviews for their constructive comments and appreciation for our work. Our responses to their key comments follow: Q1: Method innovation and clinical application (R1 & R5)：(1) For method innovation, we propose an innovative approach called the Symmetric Prior Anatomical Knowledge (SPAK)-guided GAN method. It is the first automatic surgical reconstruction method in of Orbital Blow-out Fracture (OBF) and achieves accurate and efficient orbital wall reconstruction. Despite incorporating existing frameworks, the entire procedure is novel in clinical planning. We do not feel that improved networks are always necessary, and that being able to solve clinical problems is more vital. (2) For clinical application, although the method may have limitations in bilateral orbital effects, it remains a great application prospect as unilateral orbital fractures constitute over 95% of all cases. Additionally, the supplementary video shows the successful integration of our method into the 3D Slicer software, enabling convenient and accurate automatic reconstruction of OBF, which was highly appreciated by the participating surgeons. Q2: SPAK and GAN explanations (R1, R4 & R5): Reconstruction of the orbital wall using symmetric anatomy is a common method in OBF, however existing methods are time-consuming manual operations, hence we proposed SPAK, a novel automatic generation method. It combines segmentation networks and spatial transformations, using ICP to register the 3D objects acquired from the segmentation results to obtain the deformation field. To achieve the best SPAK results, the centers of gravity of the two 3D point clouds are first coincident in ICP, followed by global registration. The quantitative evaluation of the results has been shown in Table 1, and the ablation experiment further demonstrates the SPAK plays an important guiding role in OBF reconstruction. GAN uses concatenation to integrate the SPAK output with the original ROI image. We train the SPAK model separately first and then integrate it into the GAN for end-to-end training. Notably, our entire surgical reconstruction method is fully automatic. Q3: Comparative experiment (R1 & Meta-R): Our research is the first to explore the automatic reconstruction of OBF, showing superior performance in both ablation and comparative experiments. Although not compared with some latest networks such as transformer-based networks, it is still a novel approach for automatic OBF reconstruction. Moreover, transformer-based networks often encounter challenges such as extensive computational requirements and difficult training. If our paper is accepted, we would make additions to the revision. Q4: Dataset details (R4 & R5): The dataset included 68 medial wall fractures, 29 floor wall fractures, 51 fractures involving both the medial and floor walls and 2 roof wall fractures. Data annotation was conducted by cooperative surgeons utilizing a combination of clinical semi-automatic methods and manual repair. Considering the computational requirements and ease of operation, the orbital area is cropped from the entire CT, which is a common pretreatment method. Q5: Experimental results and model evaluation. (R5): The reconstruction error of our method in the non-fractured region is remarkably small, as evidenced by the observation from Fig. 3 and Fig. 4. The surface reconstruction of U-Net, V-Net, Attention U-Net, and Attention V-Net is the same as that of the proposed method. Table 1 shows that adding DN to GN+SPAK improves the results to some extent, which is a beneficial enhancement in medical image reconstruction. The likely reason is that SPAK has played an important guiding function in GN, further increasing DN promotion effect is limited. Q6: Other minor issues (R4, R5 & Meta-R): Minor issues such as figure quality, supplemental references, showing normal and fractured images, removing ASD, etc. will be addressed in the revision if the paper is accepted.
Post-rebuttal Meta-Reviews
Meta-review # 1 (Primary)
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The paper proposes a SPAK (Specific Protein Kinase)-guided GAN (Generative Adversarial Network) model for the automatic reconstruction of the orbital wall in OBF (Orbital Blowout Fracture) surgery. Applying the GAN model to reconstruct the fractured orbital wall represents a novel approach, especially given the incorporation of prior anatomical information into the model. Moreover, the proposed method will facilitate a better understanding of the desired shape of the fractured orbital wall.
Meta-review #2
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
Pros:
- Novelty: The application of the GAN model to reconstruct fractured orbital wall is novel
- Datasets: Authors clearly described the how the dataset was collected, including the quality of the CT scans and implementation details. Cons:
- Experiments: The methods for comparison are outdated, and how they are used for evaluation is not detailed described.
- Clinical Limitation: This work is not applicable when both orbital are fractured, resulting in limited clinical feasibility. After Rebuttal:
- major issues are well explained and acknowledged;
- one unconfident reviewer reduces the score from 8 to 7, but the final scores are still high and relatively consistent
Meta-review #3
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
While this work is outside of my usual field, I did find the paper interesting. The SPAK model presented offers an interesting approach to a challenging clinical problem and in their rebuttal the authors note that surgeons appreciated the ease of incorporation with slicer. I found the ablation studies to be well done as noted by R4 with a convincing breakdown of SPAK contribution.
I lean toward accept given A) the conviction of R4 in their recommendation with clear explanation of why they are recommending accept and B) updated score of R5 leaning toward accept.