List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Rong Zhou, Houliang Zhou, Brian Y. Chen, Li Shen, Yu Zhang, Lifang He
Abstract
Integration of imaging genetics data provides unprecedented opportunities for revealing biological mechanisms underpinning diseases and certain phenotypes. In this paper, a new model called attentive deep canonical correlation analysis (ADCCA) is proposed for the diagnosis of Alzheimer’s disease using multimodal brain imaging genetics data. ADCCA combines the strengths of deep neural networks, attention mechanisms, and canonical correlation analysis to integrate and exploit the complementary information from multiple data modalities. This leads to improved interpretability and strong multimodal feature learning ability. The ADCCA model is evaluated using the ADNI database with three imaging modalities (VBM-MRI, FDG-PET, and AV45-PET) and genetic SNP data. The results indicate that this approach can achieve outstanding performance and identify meaningful biomarkers for Alzheimer’s disease diagnosis. To promote reproducibility, the code has been made publicly available at https://github.com/rongzhou7/ADCCA.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43895-0_64
SharedIt: https://rdcu.be/dnwzw
Link to the code repository
https://github.com/rongzhou7/ADCCA
Link to the dataset(s)
N/A
Reviews
Review #2
- Please describe the contribution of the paper
The paper describes a method to do some form of canonical correlation analysis on imaging and genetics data.
- 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 is pretty clear and it focuses on an interesting problem, which is identifying biomarkers for Alzheimer’s disease. The authors compare to other approaches, which is great.
Fig 1. is well done.
- 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.
I think the main weakness is that it is hard to interpret the results. Could you use this for discovery of new biomarkers? How are the biomarkers used by the method interrelated? Can this tell us anything about the biology of Alzheimer’s?
- 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 authors marked yes to everything. After reading the paper I think this is obviously not correct.
- 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
The setup of the paper is good, and there is obviously attention to detail. The aim is to aggregate information from brain imaging data and genetics SNPs to do classification.
1) Add the abbreviations in Table 1 to the caption.
2) Fig 4 is not a very good visualisation of this data. I would avoid using 3D plots.
3) In the conclusions you state: “In an exploratory analysis, we further show that the biomarkers identified by our model are closely associated with deficits in Alzheimer’s disease.” Are they not just associated with Alzheimer’s disease? What do you mean by deficits?
- 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 clear and to the point. There is minor novelty, but the results are well presented.
- 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 authors proposed a method called Attentive Deep Canonical Correlation Analysis (ADCCA) for the diagnosis of Alzheimer’s disease using multimodal imaging genetics. ADCCA is essentially an improvement over SDGCCA, combining the SDGCCA method with deep learning self-attention, to simultaneously consider the correlations between different views and within the same view in a low-dimensional space, providing a more comprehensive modeling of the relationships in multimodal data. Extensive experiments on a real ADNI dataset containing three imaging modalities (VBM-MRI, FDG-PET, and AV45-PET) and genetic SNP data demonstrated the advanced performance of the proposed model and explained how it could reveal disease-specific biomarkers consistent with neuroscience findings.
- 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.
Simplicity: ADCCA is a simple improvement over SDGCCA, using self-attention features instead of directly extracted features from DNN and modeling both within- and between-modality correlations. Interpretability: The combination of self-attention and SDGCCA provides strong interpretability compared to using multimodal deep learning methods. Stability and reproducibility: The authors validated the stability of the proposed model using 5-fold cross-validation, 4 evaluation metrics (ACC, F1, AUC, MCC), and the standard deviation of their scores. They also released the code as open-source.
- 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.
Limited novelty: Compared to SDGCCA, the proposed method only made minor modifications. Limited biomarker discovery: Although the model’s classification results seem good, the improvement in accuracy did not reflect on new biomarkers (the conclusion on interpretable biomarkers was based on literature from 2011 and 2019). Limited discussion: There is a lack of discussion and validation regarding whether the performance improvement and biomarker discovery on one dataset can be generalized to datasets of the same type.
- 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
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/2023/en/REVIEWER-GUIDELINES.html
Applying the latest research on unsupervised learning (etc. contrastive learning) to this field and improving and comparing them would be highly anticipated.
- 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?
minor improvements to SDGCAA
- 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 #1
- Please describe the contribution of the paper
The paper proposes a new model called attentive deep canonical correlation analysis (ADCCA) for diagnosing Alzheimer’s disease using brain imaging genetics data. The model combines deep neural networks, attention mechanisms, and canonical correlation analysis to integrate and exploit the complementary information from multiple data modalities. This leads to improved interpretability and strong multimodal feature learning ability. The ADCCA model is evaluated using the ADNI database with three imaging modalities (VBM-MRI, FDG-PET, and AV45-PET) and genetic SNP data. The results indicate that this approach can achieve outstanding performance and identify meaningful biomarkers for Alzheimer’s disease diagnosis.
- 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 proposed ADCCA model is a novel formulation that combines deep neural networks, attention mechanisms, and canonical correlation analysis to integrate and exploit the complementary information from multiple data modalities.
- The model has strong multimodal feature learning ability and improved interpretability, which can help identify meaningful biomarkers for Alzheimer’s disease diagnosis.
- The evaluation of the ADCCA model using the ADNI database with three imaging modalities (VBM-MRI, FDG-PET, and AV45-PET) and genetic SNP data shows outstanding performance for classifying AD vs. HC, AD vs. MCI, and MCI vs. HC groups.
- The paper demonstrates the clinical feasibility of using brain imaging genetics data for Alzheimer’s disease diagnosis, which could have a significant impact on clinical workflows and evaluations.
- 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.
- This paper only uses one single database (ADNI) for evaluation, which may limit the generalizability of the results to another popular AD dataset (e.g., AIBL).
- The paper does not compare the proposed ADCCA model with other state-of-the-art for AD diagnosis (with code/pre-trained model public), which could provide a more comprehensive evaluation of its performance. Following methods for example: Zhang S, Chen X, Ren B, et al. ,3D Global Fourier Network for Alzheimer’s Disease Diagnosis Using Structural MRI, in MICCAI 2022. Chunfeng Lian, Mingxia Liu, Jun Zhang, and Dinggang Shen. Hierarchical Fully Convolutional Network for Joint Atrophy Localization and Alzheimer’s Disease Diagnosis using Structural MRI. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020. Though they are not the multi-modal methods, they show really good performance on AD diagnosis.
- 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
Authors have provided detailed descriptions of the methods used, the data preprocessing steps, and the evaluation metrics used. Additionally, they have made the code for the ADCCA model publicly available, which promotes reproducibility and allows other researchers to build on this work. Overall, the paper appears to have taken reproducibility seriously and has made efforts to ensure that their work can be replicated by others.
- 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
See weakness.
- 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?
Though this paper provides an interesting method for AD diagnosis, the description for method should be improved and more experiments need to evaluate the performance.
- 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.
Summary
- The research presents attentive deep canonical correlation analysis (ADCCA) for brain imaging genetics data-based Alzheimer’s disease diagnosis. Deep neural networks, attention mechanisms, and canonical correlation analyses integrate and use complimentary input from many data modalities. Interpretability and multimodal feature learning increase. The ADNI database including VBM-MRI, FDG-PET imaging and genetic SNP data evaluates the ADCCA model. This method performs well and finds relevant biomarkers for Alzheimer’s disease diagnosis.
Strength
- The method is more interpretable than some similar methods addressing the same problem
Weakness
- Evaluation in only on one dataset which may limit the generalizability of the results to another popular AD dataset (e.g., AIBL).
- Comarison with other methods is not sufficient
- Some of the visualization can be improved as pointed out by R2
Author Feedback
We thank all the reviewers for their thoughtful comments and suggestions. We agree with most of the remarks and will incorporate them in the final version of the paper. In the following, we would like to take this opportunity to clarify a few points raised by the reviewers:
To Meta-Reviews: Please see Q1, Q2 and Q6 for our response to the issue of confounder.
Q1: Evaluation on only one dataset (#R1) We acknowledge that our study focused on data from a single database, specifically the ADNI cohort. By focusing on a single dataset, we aimed to ensure the consistency and reliability of our analyses, controlling for potential confounding factors that could arise from combining multiple datasets with inherent variability. Nevertheless, we acknowledge that the generalizability of our findings to other datasets, including AIBL, should be further investigated. We plan to address this limitation in future research by expanding our analysis to include additional datasets.
Q2: Comparison with other state-of-the-art methods for AD diagnosis (#R1, #R3) We understand that many of these state-of-the-art models focus on single-modal data, while our approach utilizes multi-modal data, which makes direct comparisons challenging. However, we agree that it is important to provide a more extensive comparison. In our revised version, we will make an effort to include comparative analyses with these models, acknowledging the differences in data modalities.
Q3: Interpretability (#R2) To enhance interpretability, we used the integrated gradients interface provided by Captum to assign importance scores to each feature (i.e., ROIs and SNPs) in our pre-trained model. The features with the highest scores were considered the most discriminative ROIs and SNPs related to Alzheimer’s disease (AD). This approach helps explain how input features contribute to the model’s output and provides insights into the biological relevance of these features in AD.
Q4: Reproducibility checklist (#R2) Upon reviewing the checklist again, we acknowledge that there were certain aspects that we did not explicitly address, such as providing information on the average runtime for each result or the estimated energy cost. We appreciate your observation, and in the final version of the paper, we will make sure to include the criteria outlined in the checklist.
Q5: Abbreviations (#R2) Thank you for bringing this up. We will ensure that the abbreviations used in Table 1 are included in the caption in the final version.
Q6: Visualization about 3D plots (#R2) Thank you for your feedback regarding Fig. 4 and the use of 3D plots. We agree that 3D plots may not effectively represent the data in the best possible way, although they are frequently used in parameter sensitivity analysis. As a response, we have investigated the implementation of 2D heatmap plots to visualize these multi-dimensional results, as they are more effective in conveying the information. We will substitute the 3D plots with these improved results in the final version.
Q7: Term “deficits” (#R2) Sorry for the confusion. We indeed mean that these biomarkers are directly related to Alzheimer’s disease. The term “deficits” does refer to the various impairments and abnormalities in Alzheimer’s disease. We will rephrase the statement as “In an exploratory analysis, we further show that the biomarkers identified by our model are closely associated with Alzheimer’s disease” to provide a clearer statement in the final version.
Q8: Biomarker discovery (#R3) In our study, we focused on showing the most frequently selected SNPs with high importance scores, as depicted in Fig. 3. However, we understand that the selection of SNPs was limited, and unfortunately, there is no single method capable of effectively identifying all relevant SNPs. In the future, we will explore the possibility of conducting experiments with a broader range of SNPs to potentially discover more new biomarkers.