List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Shouyue Liu, Jinkui Hao, Yanwu Xu, Huazhu Fu, Xinyu Guo, Jiang Liu, Yalin Zheng, Yonghuai Liu, Jiong Zhang, Yitian Zhao
Abstract
Optical Coherence Tomography Angiography (OCTA) is a promising tool for detecting Alzheimer’s disease (AD) by imaging the retinal microvasculature. Ophthalmologists commonly use region-based analysis, such as the ETDRS grid, to study OCTA image biomarkers and understand the correlation with AD. However, existing studies have used general deep computer vision methods, which present challenges in providing interpretable results and leveraging clinical prior knowledge. To address these challenges, we propose a novel deep-learning framework called Polar-Net. Our approach involves mapping OCTA images from Cartesian coordinates to polar coordinates, which allows for the use of approximate sector convolution and enables the implementation of the ETDRS grid-based regional analysis method commonly used in clinical practice. Furthermore, Polar-Net incorporates clinical prior information of each sector region into the training process, which further enhances its performance. Additionally, our framework adapts to acquire the importance of the corresponding retinal region, which helps researchers and clinicians understand the model’s decision-making process in detecting AD and assess its conformity to clinical observations. Through evaluations on private and public datasets, we have demonstrated that Polar-Net outperforms existing state-of-the-art methods and provides more valuable pathological evidence for the association between retinal vascular changes and AD. In addition, we also show that the two innovative modules introduced in our framework have a significant impact on improving overall performance.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43990-2_57
SharedIt: https://rdcu.be/dnwMc
Link to the code repository
https://github.com/iMED-Lab/Polar-Net-Pytorch.git
Link to the dataset(s)
https://ieee-dataport.org/open-access/octa-500
Reviews
Review #1
- Please describe the contribution of the paper
A novel approach for analyzing OCTA projections is proposed for Alzheimer detection from the eyes. The proposed approach first maps regions around the foveal zone to the polar coordinate system which enables the incorporation of the ETDRS grid, that has been proposed for focal treatment of diabetic retinopathy, as a prior knowledge about the region importance. As a result, the authors achieve more granular interpretability of the region importance in the prediction than the existing work in literature (gradCAM maps). All the components are combined in a common architecture, the Polar-Net, that can be used both for detection and improved model interpretability. The method is compared against standard and recent methods, like ResNets, ConvNext, Visual/Swin Transformers, and it outperforms them considerably in the AZ detection. In terms of interpretability, the assessment found that the model assigns similar importance to regions that clinical studies had found that are important in AD.
- 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) Methodology: The authors realize a finer and more clinically relevant interpretability method than existing work in the literature. While existing work in the literature is based on class activation maps derived from the model, it can only provide a low-resolution interpretable visualization. However, here the method incorporates the ETDRS grid to the model training, and so the algorithm is sensitized and can indicate the significance of regions in a more clinical friendly way, and also in a standard anatomical notation (e.g. importance of NI, SI, II sectors). Potentially, the method can be extended and allow the study of the influence of different sectors across different retinal layers and eyes (left vs. right) to the development and detection of ocular diseases. All the previous were made possible by transforming the regions around the FAZ to the polar coordinates system.
2) Validation and Results: The method provides a strong validation and was tested in two datasets. One in-house dataset was used that is several times larger a second public dataset. The results show that the performance of the Polar-Net is significantly higher than the existing work. I am positively surprised by the performance of the method against a wide range of methods, CNN and transformer-based methods. Especially for Kappa metric the method outperforms the existing work by 10%. The authors took into account the inter-eye correlation, and they provide confidence metrics (standard deviation). Also, they compare the interpertability against clinical findings, and they confirm that certain areas are affected disproportionately in Alzheimer patients against healthy subjects.
- 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) There are some clarity issues that might affect the reproducibility and the experimental setting. VAFF-Net was used to find the center of the fovea. The model segments the whole FAZ area and not the center. It is not clear how the authors identified the center given that the segmentation mask does not have a circular area. Also, the authors do not expand if they took confounding factors into account that can affect the vascular characteristics of the FAZ. For example, they do not provide the diabetic status of the patients. Also, according to [1] FAZ area has numerous limitations when considered as a potential biomarker for Alzheimer disease detection. For example, some studies have shown that age, gender and eyeball axial length can affect the FAZ area. Also age plays an important role and is a risk factor of AD, if you have heathly young subjects versus old diseased subjects the model might learn to identify the age group of the subjects (young vs. old) instead of their AD status. The authors do not provide the severity of AD of the patients in their dataset. 2) The authors do not discuss the limitations of their approach as well as the future direction
[1] “Variability of foveal avascular zone metrics derived from optical coherence tomograpy angiography images”, Transl. Vis. Sci. Technol. 2018
- 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
Some components of paper are available, like the FAZ detection model, while others not, like the dataset. If the paper is accepted I am sure the method could act as a reference, and be replicated by the community.
- 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
1) Why the authors considered only the area around the FAZ, and not the center of the FAZ for feature extraction? There might be useful features from the projections that could make the model more robust to diseases that affect the fovea, for example macula edema from diabetes, hypertension, or age related macula degeneration.
2) In Figure 1 the Oc and R are given did the authors normalize for different eyeball parameters?
3) At the end of the introduction section, please remove the point on releasing the code after accepting the paper, or even better please share the dataset and the code. It could be very beneficial for the community to have access to the reference implementation and dataset.
4) Figure 3 needs a bit better separation between each module, probably a grid to separate the different sections could help. Also, the first part is confusing because while the method is based on 3 input projections it shows that it can accept N number of projections. Instead, in Figure 1 the Polar-Net with its 3 PFEM modules is very clear.
5) What is the inference time of the authors’ method and compared to the existing work? Since the approach is clinically friendly, could the method fit to the time constraints of the clinical workflow?
6) In the data description, why there is such a difference between the number of patients and the number of images? Did the author discard any images because of quality issues? Also, the stage of the AD patients are not given.
7) In the tables, please include that the second number is the standard deviation.
8) In the [24] citation of the paper, which is used to assess the interpretability, it is concluded that there are changes in the layers and some parameters of the vessels, however it is not very specific to regions within a single OCTA layer. The most relevant information that I found is that in [24] there is larger roundness of FAZ in the SVC layer in mild cognitive impaired patients. However, in the evaluation of the interpretability of this paper the author claim that there are differences between sectors within a single layer. Why there is such a difference in the interpretation?
- 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 authors propose a smart way to improve the interpretability in a clinical relevant way. The results are very convincing.
- 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 #2
- Please describe the contribution of the paper
The present study introduces a novel neural network, referred to as Polar-Net, designed to detect Alzheimer’s disease via analysis of OCTA images. Specifically, the OCTA images undergo a conversion process to a polar coordinate system to facilitate the analysis of sector and ring-shaped areas which hold varying levels of significance in a clinical context.
- 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 proposed approach includes a polar transformation to better align with clinical practices. 2) A carefully designed network architecture incorporating atrous convolution, multi-kernel pooling, and CBAM attention mechanisms was employed. 3) Comprehensive experiments were conducted on both public and private datasets, and the resulting performance is thoroughly demonstrated.
- 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 incorporation of rotation operations in data augmentation may lead to ambiguity in weight allocation across different region divisions, which could potentially explain the limited performance improvement observed when incorporating prior knowledge.
- 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 promise to release the codes.
- 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
1) It is recommended that Figure 3 be reorganized to provide a clearer illustration of the network architecture. 2) In Table 2, specifically in the column labeled ACC for the MUCO-Net method, the w/o trans model outperforms the w trans model.
- 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 clinic value
- 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
This paper proposes a Polar-Net that maps OCTA images from Cartesian coordinates to polar coordinates, which allows the use of approximate sector convolution and enables the implementation of ETDRS grid-based region analysis methods commonly used in clinical practice. Polar-Net incorporates clinical prior information into training, and it is more helpful for researchers and clinicians to understand the decision-making process of the model in detecting AD and to assess its consistency with clinical observations. Through the evaluation of private and public datasets, Polar-Net proved to be superior to the existing state-of-the-art 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.
The paper addresses a very clinically relevant problem, is clearly written, conducts a thorough literature study, and clearly mentions several proposed novelties.
- 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 corresponding content for the importance map in Fig. 3.
- In Implementation details, why did the authors choose nearest-neighbor interpolation? Have other interpolation methods been tried?
- How does the weight matrix update?
- This paper mentions ‘prior knowledge’, what exactly does it contain and how is it utilized?
- In Fig. 4, the authors may consider switching to displaying the importance maps overlaid on the original image, like (b) in Fig. 2.
- 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
Upon release of the code, the reproducibility can be verified.
- 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
Comments mentioned clearly in the strength and the weakness section of the papers.
- 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 relevant weakness and comments are mentioned before but those are mostly minor and can be addressed quickly. The reviewer believes that the problems addressed in the paper is very clinically relevant and important.
- 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.
This paper proposes a polarnet to analyse OCTA projections for Alzheimer detection from the eyes in polar coordinates, which allows for the use of aproximate sector convolution and ETDRS gtride-based region analysis. Based on the three positive reviews, I recommend accepting this submission. However, there are some concerns arised by the reviewers, please revise those in the published version.
Author Feedback
Thanks for the AC and reviewers efforts, we will make further improvement in our future works.