List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Xingyue Wang, Heng Li, Zunjie Xiao, Huazhu Fu, Yitian Zhao, Richu Jin, Shuting Zhang, William Robert Kwapong, Ziyi Zhang, Hanpei Miao, Jiang Liu
Abstract
It has been suggested that the retinal vasculature alternations are associated with dementia in recent clinical studies, and the eye examination may facilitate the early screening of dementia. Optical Coherence Tomography Angiography (OCTA) has shown its superiority in visualizing superficial vascular complex (SVC), deep vascular complex (DVC), and choriocapillaris, and it has been extensively used in clinical practice. However, the information in OCTA is far from fully mined by existing methods, which straightforwardly analyze the multiple projections of OCTA by average or concatenation. These methods do not take into account the relationship between multiple projections. Accordingly, a Multi-projection Consistency and complementarity Learning Network (MUCO-Net) is proposed in this paper to explore the diagnosis of dementia based on OCTA. Firstly, a consistency and complementarity attention (CsCp) module is developed to understand the complex relationships among various projections. Then, a cross-view fusion (CVF) module is introduced to combine the multi-scale features from the CsCp. In addition, the number of input flows of the proposed modules is flexible to boost the interactions across the features from different projections. In the experiment, MUCO-Net is implemented on two OCTA datasets to screen for dementia and diagnose fundus diseases. The effectiveness of MUCO-Net is demonstrated by its superior performance to state-of-the-art methods.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16434-7_66
SharedIt: https://rdcu.be/cVRsy
Link to the code repository
N/A
Link to the dataset(s)
https://ieee-dataport.org/open-access/octa-500
Reviews
Review #2
- Please describe the contribution of the paper
In order to predict dementia from OCTA images, authors proposed CsCp module to abstract the consistent and complement representations from multiple projections of OCTA. The proposed CVF is connected after that for the feature fusion. The experiments conducted on a private dataset show its superior performance over SOTA multi-view fusion frameworks.
- 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.
- To predict dementia from OCTA is attractive, less-explored and of great clinical value.
- Authors adapted the recently popular scale dot-product attention machnism to fuse the multi-view features of OCTA, which is a good and novel application of deep learning techniques.
- Experimental results show the proposed method outperforms SOTA multi-view fusion models.
- 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.
My main concern is about the experiments. First, authors conducted the dementia prediction experiments on a private dataset. It may cause the over-estimation of the proposed method, since the hyper-parameters, training settings and data preprocessing can be specifically desinged for the dataset. Second, there is no ablation study in the experiments. I think a detailed ablation study is needed for this paper. For example, the effectiveness of consistency attention, complementarity attention, CVF, three different supervisions should be verified. In addition, authors should also show the effeciency of the model. For example, FLOPS/model size/memory-usage information are recommended to be reported.
- 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 method seems reproducible. However, the main experiments are conducted on a private dataset, which undermines the 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
- I recommend to replace section 3.2 by ablation study, authors can put extented experiment to supplyment materials
- Authors should provide some examples of different projections of OCTA.
- Better to provide FLOPS/model size/memory-usage information
- It would be more persuasive if the dataset can be publicly avaliable
- 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 article discussed an interesting topic and provided a well-designed framework to show the feasibility. The proposed method adapted the popular dot-product attention to combine multiple projection features of OCTA, which I think is a good application. However, the experiments are not sufficient to prove the effectiveness of the method.(see point 5 for details)
- Number of papers in your stack
4
- 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
Review #3
- Please describe the contribution of the paper
This paper proposes a multi-projection consensus and complementarity learning network (MUCO-Net) for dementia screening on Optical Coherence Tomography Angiography images. The proposed method performs very well on their private dataset and open OCTA-500 Dataset.
- 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 specifically solves the problem of dementia screening on Optical Coherence Tomography Angiography images. The complementarity and multi-projection fusion proposed in this paper is rarely studied in this problem and is an advantage.
- This article is well organized. In this paper, the background of the problem and the deficiencies of the previous work are clearly explained, and the corresponding modules are designed. There are certain comparative experiments and ablation experiments to illustrate 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.
- There are many hyperparameters in the loss function. The value of the hyperparameters is directly given in the paper, and there is no experiment to illustrate the influence of different hyperparameters. And the settings of the hyperparameters are not explained in the second set of experiments.
- The attention mechanism designed in the paper is not novel enough. The consistent and complementary attention module is similar to the previous positive and negative attention mechanism, and the cross-view fusion mechanism is similar to the co-attention mechanism.
- 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 credible.
- 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 authors should provide experiments on hyperparameters to demonstrate the reliability of the selected parameters.
- 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 paper is well organized, and the proposed method is innovative in dementia screening on Optical Coherence Tomography Angiography images. However, the proposed method is not novel enough.
- 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 #4
- Please describe the contribution of the paper
This paper proposes MUCO-Net to explore OCTA-based dementia diagnosis. MUCO-Net includes a consistency and complementarity attention module and a cross-view fusion module for understanding projective relationships and combining features. The method achieves the best results on two 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.
(1) The idea about Consistency and Complementarity in Screening of Dementia is interesting. (2) The experimental results in Table 1 are very good. (3) The structure of this paper is clear.
- 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.
(1) Novelty may be limited. Residual self-attention in Cross-view fusion module is similar to [1][2][3]. [1]. Exploring Self-attention for Image Recognition [2]. An Explainable 3D Residual Self-Attention Deep Neural Network For Joint Atrophy Localization and Alzheimer’s Disease Diagnosis using Structural MRI [3]. Studying the Effects of Self-Attention for Medical Image Analysis
(2) The author said that CsCp was gradually added after each feature extraction stage, and it was found that the classification accuracy became better. But why is the result of MUCO-stage3 much lower than other stages? Why is the effect of MUCO-stage1 better than other stages?
(3) In Table 2, the result improvement is not very large. Besides, are the model size and test speed inferior to the SOTA approachs?
- 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 authors will provide code after the paper is accepted.
- 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
Some minor typos should be noted, e.g., epochs is 200-> The number of epochs is 200. The experimental content is not rich enough, and the analysis should be further strengthened.
- 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?
Idea and results are good, and the task makes sense
- 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
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.
This paper presented a Multi-projection Consistency and complementarity Learning Network (MUCO-Net) dementia screening using Optical Coherence Tomography Angiography (OCTA) images. Also, a cross-view fusion (CVF) module is developed to enhance the understanding of projective relationships and combining features. The experiments on their private dataset and public OCTA-500 Dataset demonstrate the proposed method performs superior performance over SOTA multi-view fusion frameworks. The authors should revise the manuscript according to the comments of reviewers before acceptance.
- 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).
3
Author Feedback
Thank you very much for reviewing this manuscript. We appreciate all your comments and suggestions! Please check our itemized responses below.
To reviewer #2:
- We will consider the suggestions and add the results of some experiments with hyper-parameters settings in future work but did not include them in the text due to lack of space. In addition, our method is validated on an open dataset.
- Thank you for your suggestion on the ablation experiment. We subsequently added the experiment on consistency attention, and the result is 84.00(0.02)/83.18(0.03)/56.07(0.50). However, as the complementarity attention is designed based on consistency attention, the CVF module connects the results of the CsCp module, and the ablation experiments have not been set up here.
- Following your advice, we calculated them and FLOPs is 439.97G, Model size is 275M, and memory usage is 1768.92MB.
To reviewer #3:
- Thank you for your suggestion, and we will consider the suggestions and add the results of some experiments with hyper-parameters settings in future work but did not include them in the text due to lack of space.
- Our work explored a new application using OCTA to screen for dementia. In addition, we redesign and integrate the advantages of different attention mechanisms and apply them to this new field. Thank you for your suggestions. We will carry out more innovative methods of research in the future.
To reviewer #4:
- Thanks very much for the comments about some small writing mistakes and related research recommendations.
- Your suggestion is very enlightening, which will probably focus on our future work. In my perspective, MUCO-stage1 learns the lowest-level features from different projections at the first layer. In this case, the feature contains spatial information, and the consistency complementarity features in the image can be mined according to the spatial information. As the network goes deeper, spatial information is progressively reduced, leading to a decline in results.
- In Table 2, the result is relatively high because SOTA adopts the multi-task method and performs the classification task and segmentation task simultaneously. However, the multi-task method requires additional split labels, which are not required in our approach.