List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Yuanhong Chen, Hu Wang, Chong Wang, Yu Tian, Fengbei Liu, Yuyuan Liu, Michael Elliott, Davis J. McCarthy, Helen Frazer, Gustavo Carneiro
Abstract
When analysing screening mammograms, radiologists can naturally process information across two ipsilateral views of each breast, namely the cranio-caudal (CC) and mediolateral-oblique (MLO) views. These multiple related images provide complementary diagnostic information and can improve the radiologist’s classification accuracy. Unfortunately, most existing deep learning systems, trained with globally-labelled images, lack the ability to jointly analyse and integrate global and local information from these multiple views. By ignoring the potentially valuable information present in multiple images of a screening episode, one limits the potential accuracy of these systems. Here, we propose a new multi-view global-local analysis method that mimics the radiologist’s reading procedure, based on a global consistency learning and local co-occurrence learning of ipsilateral views in mammograms. Extensive experiments show that our model outperforms competing methods, in terms of classification accuracy and generalisation, on a large-scale private dataset and two publicly available datasets, where models are exclusively trained and tested with global labels.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16437-8_1
SharedIt: https://rdcu.be/cVRsN
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
The paper cover multi-view detection of mammographic abnormalities. The method uses local and global information based on standard radiology assessment. Evaluation is based on three different datasets and uses global labels.
- 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.
There is novelty in the local aspects and the specific application area.
- 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.
Lack of repeated n-fold cross validation.
Lack of discussion on the performance.
- 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
Okay.
- 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
Would swapping the CC and MLO provide the same results? It seems only the projection aspect might make a difference.
Could this be translated to other application areas. It seems most other applications don’t have similar views, so the application area might be narrow.
Could full n-fold cross validation results be included. I would suggest that the term “significant” is only used if relevant p-values indicate this.
It would be good to see a detailed discussion on why the authors think their model preforms that much better that existing SOTA approaches.
- 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?
Novel aspects, narrow application area, discussion could be more extensive.
- 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
The paper introduces the Multi-View local Co-occurrence and global Consistency Learning (MVCCL) to consider features from ipsilateral mammographic views. The authors introduced Global Consistency module penalizing the differences in feature representation from the two views and also local co-occurrence learning module that uses multi-head attention to produce a representation based on the estimation of the relationship between local regions from the two views.
- 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 of using both view of mammogram through global Consistency and local occurrence is interesting and methodologically sound. 2) The authors proposed using global consistency module to enforce similar representation of both view of mammogram is very applicable since the important features for diagnosis should be available in embedding of both views. 3) The multi-head attention modules is correctly used to derive interactions between samples from both views. 4) The authors provided good comparisons and ablation study to show the performance of their model.
- 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) In table 2, the first two columns are named “endtoend” where there is no citation and I am not sure what the authors are referring to. 2) In equation 4, the authors didn’t specify The query, Key and value vectors and left them in general format. I suggest that the authors mention specify the Q, K, V vectors. 3) There are a lot of grammatical errors such as generalisation-> generalization right hand size of the image -> right hand side optimiser -> optimizer area under the precision-recall curve (AUC-PR) -> the area …. 4) Provide additional details regarding the implementation of the model so the model can be reproduced.
- 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 declined to provide the code and the value for hyperparameters is not mentioned in the paper so the result is not 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
1) Provide additional details regarding the implementation of the model so the model can be reproduced. 2) Fix the problem mentioned in the weakness 3) There are a lot of grammatical errors such as generalisation-> generalization right hand size of the image -> right hand side optimiser -> optimizer area under the precision-recall curve (AUC-PR) -> the area ….
- 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 novel and provide extensive comparisons showing the superiority of the proposed model in comparison with the previous works.
- Number of papers in your stack
4
- 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
This paper presents a multi-view local co-occurrence and global consistency learning for mammogram classification generalization. It proposes global consistency module and multi-view local co-occurrence module to aggregate the information from two views of a breast. The generalization of the proposed approach was evaluated on four datasets, including 1) testing subset of ANON 1, 2) ANON 2, 3) CMMD, and 4) InBreast, and the results demonstrate the promising 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.
-
This study considers both local and global information from two views for mammogram classification by designing global consistency module and multi-view local co-occurrence module.
-
The generalization of the performance of the proposed approach was verified by four testing sets.
-
- 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 minor issues as follows:
-
What is FN in Fig. 1?
-
It would be better if the detailed structure of each component is provided, including that for global consistency module, multi-view local co-occurrence module, and MLP.
-
- 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 detailed structure of each component is required for reproduction.
-
Two private datasets are used in the experiments.
-
- 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
It would be interesting to demonstrate the gap of generalization. That is, for a dataset A, compare the performance of models trained by the dataset A with that trained by ANON 1.
- 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 writing is clear. The idea of using global and local information from two views is interesting. Particularly, a global consistency module and a multi-view local co-occurrence module were proposed, which are motived by the well-known transformer. The generalization of the proposed approach was evaluated on four datasets, including 1) testing subset of ANON 1, 2) ANON 2, 3) CMMD, and 4) InBreast. The experimental results demonstrate the effectiveness of the proposed approach.
- Number of papers in your stack
6
- 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
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 reports a relatively novel approach to interpreting mammography images that combines local and global views. The architecture is not fully fleshed out in the text and the diagram contains notations that are undefined (what are these FN blocks?). I recommend Accept on the condition that the authors address the issues raised by reviewers specially in terms of writing, details and clarification.
- 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).
5
Author Feedback
We thank the reviewers for their constructive review and insightful feedback. In our final version, we will provide more explanation of the unclear terms and add the missed citation. Besides, I want to clarify that the input order of the network will not affect the projection of our global consistency module (GCM) since the order of the input view is not fixed, which means the main view could be either CC or MLO. There are also two types of projection being used in GCM, namely f^(cc->mlo) and f^(mlo->cc), and the selection of the projection function is dependent on the view position of the main view. During training, along with input image and its label, we also provide the view information to assist the selection of the right projection function. In the paper, we only demonstrate one example as the main view to be CC and the auxiliary view to be MLO to simplify the presentation.