List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Chenchen Qin, Wenxue Zhou, Jianbo Chang, Yihao Chen, Dasheng Wu, Yixun Liu, Ming Feng, Renzhi Wang, Wenming Yang, Jianhua Yao
Abstract
The ideal midsagittal plane (MSP) approximately bisects the human brain into two cerebral hemispheres, and its projection on the cranial surface serves as an important guideline for surgical navigation, which lays a foundation for its significant role in assisting neurosurgeons in planning surgical incisions during preoperative planning. However, the existing plane detection algorithms are generally based on iteration procedure, which have the disadvantages of low efficiency, poor accuracy, and unable to extract the non-local plane features. In this study, we propose an end-to-end deep Hough plane network (DHPN) for ideal MSP detection, which has four highlights. First, we introduce differentiable deep Hough transform (DHT) and inverse deep Hough transform (IDHT) to achieve the mutual transformation between semantic features and Hough features, which converts and simplifies the plane detection problem in the image space into a keypoint detection problem in the Hough space. Second, we design a sparse DHT strategy to increase the sparsity of features, improving inference speed and greatly reducing calculation cost in the voting process. Third, we propose a Hough pyramid attention network (HPAN) to further extract non-local features by aggregating Hough attention modules (HAM). Fourth, we introduce dual space supervision (DSS) mechanism to integrate training loss from both image and Hough spaces. Through extensive validations on a large in-house dataset, our method outperforms state-of-the-art methods on the ideal MSP detection task.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16449-1_56
SharedIt: https://rdcu.be/cVRXu
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 deals with automated detection of the midsagittal plane (MSP) for brain surgical planning. The MSP bisects the human brain into two cerebral hemispheres. The detection of optimal MSP is based on a deep Hough plane network which combines the idea of Hough Transform based object detection with deep convolutional neural networks.
- 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 main strengths of the paper is the interesting methological approach combining hough transform and deep learning. In this way, a nice framework is being created which could also be of interest for other applications.
Another strength is the evaluation on a decent number of 519 cases providing first indication of clinical usibility.
- 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.
Reproducibility is unfortunately rather moderate (if at all). Details on runtime and memory consumption are missing although the method is being claimed to be beneficial from that perspective.
For the comparison against other approaches, it remains unclear if those are based on available implementation or re-implementation. Of course, in the latter it is uncertain if alternative methods are implemented correctly and optimally tweaked.
In general, more justification should be provided why proposed method significantly outperforms SOA.
- 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
Unfortunately, implementation of the method is not given. From that perspective, reimplementation would be needed. Provided level of detail is however limited and consequently reproducibility is only moderate.
- 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/2022/en/REVIEWER-GUIDELINES.html
In general the paper is nicely written and easy to follow. It would be great to share code and allow others to reproduce results and apply method to other use cases.
The comparison to state of the art would require more justification. From me, the conclusion can often not be fully understood, e.g., “Consequently, they commonly have the disadvantages of low efficiency, poor accuracy, […], leading to unsatisfactory and suboptimal results.” However, some of the approaches give for their application promising results and not so sure if I agree with such conclusion. Another example is “In addition, insufficient number of landmarks will also make the plane fitting unstable, which ultimately leads to the poor robustness and accuracy of the plane detection based on the location of landmarks.” Again, I can not fully follow the argumentation as a plane fit does not require a lot landmarks and I would argue that usually enough 3D landmarks can be detected. At the end, ground truths is extracted from annotated landmarks.
Figure 4 deserves further explanation. How have the cases been selected? What can be observed in detail? How can the difference be explained?
Very nice to split the data in different levels of pathology (>5mm and < (=?) 5mm)
- 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?
Certainly interesting paper with nice methodological contribution that could be relevant for other applications. Unfortunately, reproducibility is rather moderate. It is less transparent why improved results over state of the art are achieved.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
2
- 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
In the sagittal plane localization task of 3DCT human brain image:
- Make full use of image space and Hough space features to locate the sagittal plane.
- The use of DHT increases feature sparsity and reduces computational cost.
- 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 method proposed in this paper is novel: combining two kinds of spatial characteristic information into training can accomplish tasks more efficiently and accurately. This paper has high practical value: based on real clinical data sets, it has good training 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.
- There is no reasonable explanation for the selection of some parameters.
- The rationality of the method description is slightly inadequate.
- The number of comparative tests is insufficient, and the comparison method selected is not new enough.
- The types of experiments are not rich enough, lacking some ablation experiments.
- Insufficient references.
- The summary part is less.
- 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 network architecture is clear and the algorithm logic is rigorous. Only in the experimental part is not obvious, the comparison test is not full; Clinical needs studies based on real data sets; It has relatively good reproducibility.
- 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/2022/en/REVIEWER-GUIDELINES.html
- P1 Abstract: For Abstract, data types and forms are not mentioned, which can be appropriately supplemented. Let the reader have a more comprehensive understanding of the work of the paper.
- P2 line1-7: It’s not that the explanation for “manually delineated guideline” isn’t clear. The specific way of manual labeling leads to bad clinical effects. I think this part of the explanation can be modified to give readers a more intuitive feeling.
- P2 Fig.1. (a): What is the specific impact of this deviation on the operation.
- P5 Fig.3. line2-4: There is no specific explanation for this part. Can you use a small amount of space or formula to show it in the text?
- P5 2.2 (2): Such voting process can greatly improve the training efficiency, but will it affect the training effect to some extent if all pixels with a space median value of 0 are erased? Is it possible to add comparative tests to verify that the training efficiency can be improved while ensuring the improvement of recognition accuracy?
- P6 2.4 line1-2: Can you give a relatively specific explanation, or add paper citations with corresponding explanations?
- P6 2.4 λ: Can you explain the rationality of setting 0.1? Or add experiments to illustrate.
- P7 line8-10: The relationship between GT and Landmark is not clearly expressed.
- P7 line11 3:1:1: The proportion of test sets is relatively small.
- P7 3.2 Table1: The comparative tests are a little sparse. The latest comparison is 2019. Is there an updated method to compare with the method proposed in this chapter?
- P9 References: There are relatively few references.
- 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
6
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper is logical and precise, and the research content has certain clinical value. It has some bright spots, and really improves the effectiveness of the work in this task area. But there are also some problems such as unclear description, lack of explanation and inadequate experiment.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
1
- 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 #2
- Please describe the contribution of the paper
Following are the contribution of the paper:
- Application of deep hough transform (DHT) and inverse deep hough transform (IDHT) which simplifies the plane detection problem into image space.
- Combination of Hough transform along with CNN features to detect the MSP properly.
- 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.
Following are the strength of paper:
- Introduction of sparse DHT, which increases the sparsity of features, which further increases the computation speed.
- Introduction of hough pyramid attention network to extract non local features, which increases the detection.
- Introduction of dual space supervision, which integrates the training loss from both the spaces (image and hough).
- 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 work is interesting. I have no doubt on the novelty of the algorithm, but I am concerned about the application of the work. Detection of MSP looks good, the strength of the paper would increase if the midline shift could be calculated. In the Figure 1, the axial brain CT slices, the ideal midline or ideal MSP is detected, it would be better if the deformed or the midline offset is also detected. After detection of ideal and offset midline, the midline shift could be calculated, which is one of the important parameter in predicting the severity of the brain. It would be better if midline shift is calculated, and a better validation would be the error in midline shift estimation.
- 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 is reproducible.
- 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/2022/en/REVIEWER-GUIDELINES.html
The paper is well written and understood. It would be better that if the authors would detect the deformed beizer curve and estimate the midline shift. The midline shift is an improtant predictor for predicting the sevearity of the brain. The application of hough transform is interesting to see, but tracing the deformed curve is equally important.
- 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 would recommend it weak accept as the midline shift estimate is not done. With the estimation of midline shift I would recommend it strong accept.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
3
- 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 #4
- Please describe the contribution of the paper
- The authors propose to detect midsaggital plane (MSP) for brain surgical planning extending the deep hough transform network [8] to plane detection in 3D. The application is interesting and shows significant improvements over the baseline MSP detection methods.
- 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 paper with clear clinical motivation. Also good review on different MSP landmark detection methods
- Novel method extending the 2D DHT to 3D with additional modifications for an end-to-end network
- 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.
- Figures 2 need more details in caption to understand each module. At present it is not self explanatory
- How did they choose lambda to be 0.1? Any rationale behind this choice?
- 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
- Authors have selected not to release code and related dataset, thus reproducibility of this work cannot be guaranteed. While, I understand the authors reservations, I recommend them to release inference codes with trained checkpoint and a subset of dataset included in either supplementary or main paper. If this makes sense?
- 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/2022/en/REVIEWER-GUIDELINES.html
- DHT & IDHT please replace with DHT and IDHT; ReLu and not Relu; ResNet and not resnet; Qi Han et al. [8] should be Kai Zhao et al. [8]
- Please briefly summarise the Figure 2 in your methodology.
- Conclusion should include future directions. Requires rewriting, please avoid words like ‘great …’ as they are non-scientific words but rather summarise experimental-based hypothesis and findings.
- 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 paper does hold a promise and the authors have included novelty for 3D plane detection using deep hough transform. Also, added DSS and attention networks does show significant improvements.
- Number of papers in your stack
8
- What is the ranking of this paper in your review stack?
2
- 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
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 reviewers unanimously agree about the suitability of the work for presentation at the 2022 MICCAI meeting. They also suggest several revisions to improve the work. The authors should make every effort to address these in their final version.
Remaining weaknesses (meta reviewer) The authors are strongly encouraged to make the code publicly available for reproducibility purposes. The method is not compared to standard plane detection algorithms mentioned in the introduction.
In the final version of the paper please include the following information:
- runtime and memory consumption should be included
- Provide more details on how the baseline methods were implemented and trained.
- Provide more explanation on the figure captions ( Fig 2 and Fig 4 for example)
- Explain why the constant values for different parameters were set (Rev 3)
- What is the ranking of this paper in your stack? Use a number between 1 (best paper in your stack) and n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
2
Author Feedback
N/A