List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Weiyuan Lin, Hui Liu, Lin Gu, Zhifan Gao
Abstract
Morphological segmentation of the aorta is significant for aortic diagnosis, intervention, and prognosis. However, it is difficult for existing methods to achieve the continuity of spatial information and the integrity of morphological extraction, due to the gradually variable and irregular geometry of the aorta in the long-sequence computed tomography (CT). In this paper, we propose a geometry-constrained deformable attention network (GDAN) to learn the aortic common features through interaction with context information of the anatomical space. The deformable attention extractor in our model can adaptively adjust the position and the size of patches to match different shapes of the aorta. The self-attention mechanism is also helpful to explore the long-range dependency in CT sequences and capture more semantic features. The geometry-constrainted guider simplifies the morphological representation with a high spatial similarity. The guider imposes strong constraints on geometric boundaries, which changes the sensitivity of gradually variable aortic morphology in the network. Guider can assist the correct extraction of semantic features combining deformable attention extractor. In 204 cases of aortic CT dataset, including 42 normal aorta, 45 coarctation of the aorta, and 107 aortic dissection, our method obtained a mean dice similarity coefficient of 0.943 on the test set (20%), outperforming 6 state-of-the-art methods about aortic segmentation.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16443-9_28
SharedIt: https://rdcu.be/cVRyG
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 work is dedicated on the automatic segmentation of the aorta from injected CT-scan using a deep learning approach that is geometry-constrained.
- 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 this work are :
- Considering the shape of the aorta in the segmentation process, and then by extent, the fact that the segmentation must be anatomically plausible
- Validation on a database with different diseases
- Evaluation of the method according to the part of the thoracic aorta (ascending, arch and descending)
- Good results
- 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 are some weakness in this paper, however :
- Who did the manual segmentation ?
- Making an evaluation according to the part of the aorta is a good idea, but it could be interesting to have the same kinds of result according to the pathology
- The Hausdorff distance is a metric to detect the outliers, and only considering the 95% of the Hausdorff distance is a non-sense, or something to hide bad results. Please provide the “true” HD
- The standard deviation must be indicated in the Bland Altmann plots in the figure 5.
- The diameters indicated in the discussion are not good. According to the Bland Altmann plots, the maximum error is higher than 0.98 mm
- The units are not specify for the HD
- 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
It would be interesting to have access to the database, but I know that it is not evident. Moreover, the manual segmentation is questionable
- 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
My main comments to improve this article are :
- Provide the Hausdorff distance
- Indicate the results according to the diseases
- Some errors in the text, particularly the values in the discussion
- Specify the units for the Hausdorff distance and the RMSE
- 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?
Good article, good study, but some correction to do.
- Number of papers in your stack
3
- 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
The paper introduces a geometry-constrained pipeline for the automatic segmentation of the aortic lumen in healthy and diseased cases. The diseased cases include coarctations and aortic dissections. The proposed method uses geometry-constrained deformable attention which consists of an extractor and a guider. The former generates variable-size patches and, thanks to self attention, captures long-range dependencies.
- 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 work proposed a novel pipeline for the automatic segmentation of the aortic lumen. The authors show that this pipeline outperforms previous works on both healthy and diseased aortas. It is of potential interest to the community given the limited amount of work on image analysis of chronic aortic syndromes [1,2].
[1] Fleischmann, Dominik, et al. “Imaging and Surveillance of Chronic Aortic Dissection: A Scientific Statement From the American Heart Association.” Circulation: Cardiovascular Imaging 15.3 (2022): e000075. [2] Pepe, Antonio, et al. “Detection, segmentation, simulation and visualization of aortic dissections: A review.” Medical image analysis 65 (2020): 101773.
- 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 introduction is rather long for the format of the paper, but while it brings attention to some problems of cardiovascular radiology, it does not provide much information on the current state of the art, its limitations and why the reasons behind the architectural choice. Perhaps recent reviews on this topic can support the authors in editing the introductory part [1,2]. I would also recommend the authors to split the introduction from the related works.
- In 2.1, “h” is not defined. Also, what do the authors mean with “base fact”?
- While comprehensive, I find the method description (section 2) to be too focused on the mathematical description of some aspects, while other aspects are left without any details (e.g. “At the end, a lightweight decoder is used to fuse multi-level features to decode.” -> it is not clear what the decoder actually is and how it exactly works).
- Dataset: Was there an ethical approval for the access of the 204 CT scans? Please report statistics regarding volume size and spacing.
- Train and test: after how many epochs was the learning rate reduced?
- Diameter meausurements were not compared against state of the art but only against ground truth.
[1] Fleischmann, Dominik, et al. “Imaging and Surveillance of Chronic Aortic Dissection: A Scientific Statement From the American Heart Association.” Circulation: Cardiovascular Imaging 15.3 (2022): e000075. [2] Pepe, Antonio, et al. “Detection, segmentation, simulation and visualization of aortic dissections: A review.” Medical image analysis 65 (2020): 101773.
- 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
Some further implementation details are provided as supplementary materials. Do authors do not plan to publish code or trained models. CT data is also not publicly available. Details on ground truth generation are also limited.
- 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
For major points please see 4) and partly 7).
In addition:
- Minor typos/punctuation errors are present throughout the manuscript. Although minor, I would suggest the authors to read once more through the text and figure captions. E.g. sometimes it says “extracter” and some “extractor”. I assume the latter is correct?
- What do the authors mean with “reference point” in 2.2?
- Figures need to be replaced later in the text. Ex: fig.3 is mentioned much later.
- Figure 4 is currently not mentioned in the text.
- conclusions: the reported error should have a +/- sign?
- It would be beneficial to add more details about image statistics (e.g. size) and about ground truth generation protocols/processes.
- While it is meaningul to show that the proposed approach outperform other methods, it would be also interesting to show how the different methods perform on the different classes (dissection, coarctation, etc..), perhaps this could be part of a larger journal version.
- 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?
- While results show some merit, a clear comparison for diameter meausurements is currently missing
- In the current state, a solid round of editing is needed to increase readability of the manuscript.
- The editing phase should also try to clarify all aspects in the methodology section to ensure reproducibility.
- 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 #3
- Please describe the contribution of the paper
The paper proposes a geometry-constrained deformable attention network (GDAN) for segmenting aorta. There are two main contributions: morphological constraints and deformable attention mechanism are added to the segmentation network.
- 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.
Strong evalutation: The proposed method is applied to the aorta segmentation and diameter measurement. The paper gives a reasonable detailed comparison with other methods which shows the performance of the proposed 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 experimental section lacks analysis in the discussion of comparison 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
Although the data and code is not public, the authors list most of the details of the method in 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/2022/en/REVIEWER-GUIDELINES.html
The methodology is in general well described. The following comments could be taken into account for further improvements. In the experimental part, what technologies are mainly used in other methods[2,3,4,5,18.27]? Comparison should be added in related work or the experimental part to better demonstrate the advantages of the method in this paper.
In clinical application, is the vascular segmentation and diameter calculation real-time? It is suggested to increase the calculation amount compared with other methods.
There are some expressions which should be clarified, though, as they are sometimes difficult to interpret:
-
Sec 2.3: The number of center-points, M, should not be a component of C, which would make formula(5) ambiguous. M should represent the number of centerpoint information tuples. If my understanding above is correct, I think to rewrite for C^m = (l_m, v_m, d_m), C = {C^m m = 1,… M}
-
- 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 method has high clinical application value in the field of medical image analysis and also has strong experimental comparison.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
3
- 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 paper introduces a geometry-constrained pipeline for the automatic segmentation of the aortic lumen in healthy and diseased cases. The proposed method is novel and has been validated on a database with different diseases. The manuscript has minor grammar errors/typos, which needs further proofreading. The reviewers also required more experimental details, which should be addressed in the final version.
- 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
1.Analysis (R3) and limitation (R2) of comparison methods. 1)Multi-stage methods rely on stage cooperation, causing errors accumulation, such as coarse-to-fine methods. 2)Straightening can easily identify the aorta, but recovery process of the original shape causes errors, especially for curved arch. 3)Prior shape utilizes prior anatomical knowledge, but it cannot deal with complex diseases, such as aortic dissection. 4)Centerline-based CNN methods combine the distance from voxel to the boundary, but centerlines rely on manual labels, which is unsuitable for the irregular aorta in long sequences.
2.Advantages (R3) and architectural reason (R2) of our method. 1)Our geometry-constrained deformable attention method can deal with the relationship between spatial position and variable morphology in the long-sequence CT images. R1 also agrees that our work considers shape of the aorta in the segmentation process. Deformable attention module predicts offset and scale of the reference point (original center point of each patch) and finally gets new adaptive patches. Adaptive patches extract complete semantic information with long-range dependences and match aortic structures of different shapes, while fixed patches lead to separation of the intact morphology and destroy the semantic integrity. CPR-based aorta has a higher spatial similarity and homogeneity in geometry structure, which simplifies irregular and variable morphology. CPR-based images strengthen boundary constraint and improve sensitivity to geometric contours. 2)For our network, encoder extracts semantic information from raw images and contour features from CPR images, respectively, so it better realizes interaction of spatial context information between contours and masks at different resolutions. Lightweight decoder consists of pyramid-like multi-layer perceptrons, which reduces the number of parameters and computational complexity. It parses the relationship between aortic anatomy and semantics. Our method is feasible and effective.
3.For the experiment, R1 and R3 praise our evaluation. However, we add new experimental results to be more persuasive. 1)We calculate true Hausdorff distance (HD) and statistics show that the maximum error between true and 95% HD is less than 0.56mm. We have added true HD in table 1 to avoid disputes. (R1) 2)For maximum diameter meausurement, mean bias between ground truth and other methods is -2.48~ -3.62mm while ours is -1.55mm. For minimum diameter meausurement, mean bias between ground truth and other methods is -1.83~ -3.07mm while ours is -0.98mm. For different diseases, mean volume errors of CoA and dissection are 346 and 415 mm^3 in our method while they are 375~483 and 476~652 mm^3 in other methods. (R2, R1) Hence, our method outperforms other methods. 3)Our method is real-time in segmentation. It takes an average of 7.8s per case while other methods take 5.4~15.6s, ranking third. (R3)
4.Detail. (R1,R2) 1)Our work is a retrospective study. It has been exempted from formal ethical approval by the Medical Research Ethics Committees of the hospital. Besides, we supplement implementation details about dataset and train process to improve reproducibility. 2)For the dataset, the volume size is 512512Z (300≤Z≤800), and the spacing is 0.75mm. A radiologist with 6 years of experience in CT manually annotates images in a voxel-wise manner as ground truth and verifies them again. 3)For training, the initial learning rate is 0.01 and it is reduced after 5 epochs.
5.For typos/punctuation errors (R1,R2,R3), we have proofread the text and will perform in the final version. 1)For charts, the position of Fig3 is adjusted to match the text. Fig4 and standard deviation are added in Sec3 and Fig5, respectively. We add the unit (mm) in table 1 2)For formulas, we optimize description to make them clarified. “h” of formula(1) is the target space and “base fact” is the ground truth. We have corrected statement in the discussion and conclusion.