Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Disease progression prediction is a fundamental yet challenging task in neurodegenerative disorders. Despite extensive research endeavors, disease progression fitting on brain imaging data alone may yield suboptimal performance due to the effect of potential interactions between genetic variations, proteomic expressions, and environmental exposures on the disease progression. To fill this gap, we draw on the idea of mutual-assistance (MA) learning and accordingly propose a fresh and powerful scheme, referred to as Mutual-Assistance Disease Progression fitting and Genotype-by-Environment interaction identification approach (MA-DPxGE). Specifically, our model jointly performs disease progression fitting using longitudinal imaging phenotypes and identification of genotype-by-environment interaction factors. To ensure stability and interpretability, we employ innovative penalties to discern significant risk factors. Moreover, we meticulously design adaptive mechanisms for loss-term reweighting, ensuring fair adjustments for each prediction task. Furthermore, due to high-dimensional genotype-by-environment interactions, we devise a rapid and efficient strategy to reduce runtime, ensuring practical availability and applicability. Experimental results on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset reveal that MA-DPxGE demonstrates superior performance compared to state-of-the-art approaches while maintaining exceptional interpretability. This outcome is pivotal in elucidating disease progression patterns and establishing effective strategies to mitigate or halt disease advancement.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2009_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: N/A

Link to the Code Repository

N/A

Link to the Dataset(s)

https://adni.loni.usc.edu/

BibTex

@InProceedings{Zha_Disease_MICCAI2024,
        author = { Zhang, Jin and Shang, Muheng and Yang, Yan and Guo, Lei and Han, Junwei and Du, Lei},
        title = { { Disease Progression Prediction Incorporating Genotype-Environment Interactions: A Longitudinal Neurodegenerative Disorder Study } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15003},
        month = {October},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    This paper proposes a framework called MA-DPxGE for simultaneous disease progression modeling and identification of factors related to genotype-by-environment (GxE) interactions to differentiate between healthy, mild cognitively impaired and Alzheimer diseased patients. They combine this with a loss that prioritizes difficult predictions task over easy ones. They’re framework is evaluated using longitudinal imaging phenotypes at 4 time points and relevant demographic/clinical, genetic and environmental data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI).

  • 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) Divide-and-conquer strategy significantly reduced compute time while maintaining performance. 2) Inclusion of interpretablility of the interactions. 3) Well structured ablation study.

  • 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) More information is required on processing of data. Which sequences were used from imaging data. Was both FreeSurfer and VBM used for segmentation and longitudinal analysis? Why is it interesting to compare these and not just do one? Which morphometric measurements were included (volumes, surface areas, or thicknesses)? 2) Lack of cohort description.

    • Which ADNI cohort was used? Looking at this paper, there are four different original ADNI cohorts (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9402812/, Table 3).
    • What were your inclusion/exclusion criteria?
    • Additionally, why were only “non-Hispanic Caucasian participants” included? I understand that the large majority of participants in ADNI are Caucasian, however, only including these biases your model towards this population. 3) Lack of external validation or validation within a different ethnocultural group.
  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not provide sufficient information for reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    No code or data provided. As stated above, a main limitation of the current paper is the description of the dataset used and the data processing steps. This information is essential for any reproduction of analysis pipelines and results.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    • I found it difficult to understand some parts of the paper. Therefore would recommend attention be given to language correction to improve readability and clarity.

    • The current structure needs improvement. Section 3 is too much of a mixture between methods and results, although you have a dedicated methods section. Would recommend restructuring to something similar to the following:
      1. Introduction
      2. Methods 2.1 Dataset 2.2 MA-DPxGE 2.2.1 Description 2.2.2 Experiments 2.3 Evaluation
      3. Results 3.1 Comparison with state-of-the-art 3.2 Identification and Interpretation of Biomarkers 3.3 Ablation study
      4. Conclusions

    • Table 1, suggest inclusion of statistical test here to compare each step in the ablation study to baseline performance. It would be interesting to see, from which addition, the performance becomes statistically higher (if any) compared to baseline.

    • In the contribution part of the Introduction you state: “We separate the normal aging effect from the disease trajectory.” However, in Section 3.2, Interactions effects, the “intercept derived interactions include (rs8106922, age), …”. Why is age still included here?

    • Please include limitations of the current work.

    • Personally, all Figures are too small to read. Currently, I had to substantially zoom in to read the image.
    • Figure 1a and 1b) were any of the CCC and RMSE between the different methods statistically significant?

    • Figure 1c) T-test seems to be wrong statistical test here. Since you are comparing multiple groups (unpaired data), you should either use ANOVA, if assumptions for the test are met (e.g. Normality), or an alternative non-parametric test, such as the Kruskal-Wallis test, if the assumptions of ANOVA are violated. Additionally, why are the results of the Parahippocampus not shown?

    • Please sort references according to appearance in the paper, and not alphabetically.
  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making

    Weak Reject — could be rejected, dependent on rebuttal (3)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
    • Lack of evaluation of generalisability / model bias (only included data from patients from one ethnic group)
    • non-optimal description of methods and presentation of results
  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • [Post rebuttal] Please justify your decision

    Authors have responded to my suggestion and commit to modifying the paper structure and enhance data description to increase readability. Since further experiments can’t be added at this stage to assess generalisability and potential bias, this should be added as a limitation.



Review #2

  • Please describe the contribution of the paper

    The authors use mutual-assistance to model Alzheimer’s disease progression and identify informative genetic, proteomic, and environmental features as well as interactions between these features.

  • 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 authors note that these efforts may help to resolve the missing heritability of Alzheimer’s. I would agree that interactions, which have historically been difficult to systematically evaluate are likely important contributors. The results presented replicate well known and accepted associations, making the additional novel findings intriguing.

  • 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 robustness and reproducibility of the results is unclear to me. In particular, there is no clear discussion on how false discovery rate/multiple hypothesis correction is considered - something that is essential given the large number of variables considered/multi-dimensional analysis.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not provide sufficient information for reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    The logic behind the work is well presented (perhaps appropriately detailed given the short manuscript) but not in sufficient detail to reproduce.

    Data is publicly available.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
    • It is unclear to me how the proposed work gets around the statistical limitations inherent in multiple hypothesis correction. Certainly in the ANOVA testing of identified loci I would expect a multiple hypothesis correction and it is not clear this was done although the p-values presented are robust enough that I would expect them to remain significant.
    • Proteomic data is not independent of the genetic data (ie some genetic variants will affect protect levels). How does the approach consider this interdependence?
    • Some well known associations are identified. How many of the identified associations are new?
    • Although I appreciate the need for clever solutions to the computational intensity of scanning for interactions across the genome, I’m not sure I am a fan of partitioning especially when looking for interactions. It seems it would enable spurious results seen in a subset of the data to be propagated forward or miss important interactions by excluding the pairing.

    • The authors discuss the ability to prioritize challenging tasks but it is unclear to me how this is decided/titrated and what effect the weighting towards challenging tasks has on the overall outcome.

    • This sort of work, which is fantastically cool, is definitely limited by data availability. If it were available, a compare and contrast with other types of dementia that demonstrate different patterns of longitudinal progression would be interesting. I imagine there are shared features as well as those unique to AD versus FTD etc. This might also have clinical utility as it is sometimes non-trivial to diagnosis the specific subtype of dementia and as newer medications are becoming available for specific types of dementia this specific diagnosis becomes quite clinically relevant.
    • I would be hesitant to include MCI unless you know they go on to develop AD. It is a very fuzzy diagnosis at best and introducing such noise into the phenotyping while undertaking a complex, layered analysis looking for interaction terms in a relatively small dataset may introduce more noise than the increased sample size resolves.
    • It is unclear to me what is driving the better performance of this method relative to the others (Fig 1). What are the inputs for these various methods? If incorporating more data types (i.e. environmental factors) I would expect better prediction even without improved methodologic approach. Are these really comparable? Agree that considering progression is inherently better than just comparing different time points - this is what we do clinically compare time points with a disease framework in mind.
    • It would be interesting to see direction of affect for the SNP x environment interactions. For the example given about rs439401-education, is this in a direction where an intervention to increase education would help? This would be a relatively benign “treatment” with perhaps a beneficial outcome. (beyond diagnosis some of these snp x environment interactions may help in treatment with either slowing the disease or counseling family to reduce risk of others who may also carry the variant)

    Minor typographical recommendations

    • please remove the ‘s’ from the definition of “MCIs” (I presume mild cognitive impairment) and “HCs” (healthy control) used in section 3. MCI and HC are used later and it makes it difficult to search for these terms when they are not identical
    • I would use a single term to limit confusion, I presume quantitative trait = endophenotype
    • please define SVR in section 3.3. I presume it stands for “sparse variable regularizer” in the section header
  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    I like the premise of the paper, although I do think it falls victim to the “analysis is more sophisticated than the data” phenomenon in which the dataset is not sufficient in size or robustness in terms of phenotyping to warrant such an multi-layered approach. Conceptually I struggle with how much confidence can be placed on the identified interactions which is my biggest hesitation. Some discussion around false discovery could lend clarity to this.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [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

    In this paper, the authors propose a Mutual-Assistance Disease Progression fitting and Genotype-by-Environment interaction identification approach (MA-DPxGE), which can fully utilize the longitudinal imaging phenotypes and genotypes and environment and their interactions for disease progression fitting. In addition, the authors employ innovative penalties to discern significant risk factors, and carefully design loss term and adaptive mechanisms to ensure fair adjustments for each prediction task. Finally, a rapid and efficient strategy is designed to reduce the algorithm running time.

  • 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 by the authors uses genetic variations, proteomic expressions and environmental exposures to predict longitudinal imaging phenotypes. In the meantime, the authors explore the genotype-environment interaction and the genotype-protein interaction. In contrast, other current approaches have only used imaging phenotypes for modeling, and exploring the relationship between imaging phenotypes and SNPs, but have not taken into account the effects of proteomic expressions and environmental exposures. The authors set three penalty terms to achieve model sparsity while encouraging the identification of interpretable genotype-environment interactions and genotype-protein interactions. Finally, the authors design a dynamic task balancing module, which allows for adaptive weighting of loss terms. This mechanism encourages the prioritization of harder tasks among multiple prediction tasks, ensuring that they are fully optimized and preventing easier tasks from dominating.

  • 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.The authors should explain the role of the two parts, baseline status (intercept) and the changing rate (slope), as well as their relevance, otherwise it is difficult for the reader to understand why the formula is divided into two parts. The tj in formula (1) is not explained. 2.Q in the formula is the interaction term between the SNP and the environment, and why it is a separate term when analyzing genotype-protein interactions. 3.The principle and role of the FGL2,1-norm should be explained in detail. 4.The description of dynamic task balancing module is not clear enough, including adaptive mechanism and loss item reweighting.

  • Please rate the clarity and organization of this paper

    Very Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not provide sufficient information for reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    The paper as a whole is not described clearly enough and needs more explanation. For example, a detailed explanation of the settings for baseline status (intercept) and the changing rate (slope), as well as how the SNPs and genotype-environmental interactions is grouped and why grouping improves efficiency.

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making

    Accept — should be accepted, independent of rebuttal (5)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    This study proposes an interpretable framework for disease progression prediction, using genetic variations, proteomic expressions, environmental exposures and exploring the interaction between these models. In addition, the authors demonstrate correlations between genotype and proteins, as well as correlations between genotype and the environment, which have significant implications for improving the understanding of AD.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [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



Author Feedback

We thank all reviewers and are pleased that 2 of 3 recommended acceptance. We are encouraged to hear that they found significant implications (R1), incredibly cool, novel findings (R4), and interpretability (R5). Our responses to the main points are as follows: Genotype-protein correlation (GPC), genotype-age interaction, and genotype-environment interaction (R1+4+5): The GPC module extracts meaningful genotype-protein patterns because genes and proteins are highly correlated. We separate the normal aging effect (irrelevant to AD) from the disease progression trajectory, while interaction between the gene and age still exists. The effects of SNP on AD to a certain extent depend on age, thus the interaction was included in our model. The value of genotype-environment interaction can be: (1) Choosing the best treatment for the individual to maximize response or minimize side effects. (2) Predicting potential changes to modifiable environmental factors. Thus, the direction of the top identified genotype-environment interaction will be analyzed to show the effectiveness and possible treatment. Baseline status (intercept) and the changing rate (slope) (R1): Both the baseline status and the changing rate of disease progression for AD were influenced by genetic variations, and thus we used both of them in our model. MCI subjects (R4): The MCI subjects in this study are all progressive ones (finally diagnosed as AD), and stable MCI subjects were removed. Optimization and gene partition (R1+4): SNPs naturally form block structures, and the information are mainly carried by blocks, thus gene partition is reasonable for most cases. The ablation studies show that gene partition can alleviate computational challenges without sacrificing model performance. We also conducted statistical analysis and post-analysis, and all p-values reached a significance level after correction. Dynamic task prioritization (R1+R4): DTP adaptively adjusts the weights of each sub-task to automatically prioritize more challenging sub-tasks. The dynamic task prioritization technique can ensure an overall optimal and this will be clarified. Comparisons and statistical analysis (R4): All methods worked on the same inputs, and thus the performance gain was not only due to the data but also our newly designed method by combining disease progression and biomarker identification jointly. The Bonferroni correction is used to ensure robustness for multiple comparison issues in Section 3.2. We will add p values compared to the baseline method in terms of CCC and RMSE in Table 1. In Figure 1 (c), we compared every pair of groups rather than three groups (HCs, MCIs, ADs) together. The parahippocampus’ results were similar to the hippocampus and were omitted due to space limitations. Organization, dataset description, and preprocessing (R5): We will rearrange our paper as suggested by R5. The dataset description, preprocessing, and discussion with limitations will be spun off into separate paragraphs. Further generalizability (R4+5): MA-DPxGE was evaluated by two different imaging data types (the grey matter density and surface area), which could demonstrate the good adaptability of our method. We also have conducted quantitative evaluation via a simulation study but only real results were presented due to space limitations. Additionally, we are working on additional ethnic groups to further demonstrate its generalizability. Reproducibility (R4+5): The code will be released if accepted. Figure fonts and typos (R4+R5): We will enlarge the fonts of all figures, and fix all typos as suggested.



Meta-Review

Meta-review #1

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    N/A

  • What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/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).

    N/A



Meta-review #2

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    Novel interpretable framework that takes into account proteomic expressions and environmental exposures

  • What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/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).

    Novel interpretable framework that takes into account proteomic expressions and environmental exposures



back to top