List of Papers Browse by Subject Areas Author List
Abstract
Neurodevelopment is exceptionally dynamic and critical during infancy, as many neurodevelopmental disorders emerge from abnormal brain development during this stage. Obtaining a full trajectory of neurodevelopment from existing incomplete longitudinal data can enrich our limited understanding of normal early brain development and help identify neurodevelopmental disorders. Although many regression models and deep learning methods have been proposed for longitudinal prediction based on incomplete datasets, they have two major drawbacks. First, regression models suffered from the strict requirements of input and output time points, which is less useful in practical scenarios. Second, although existing deep learning methods could predict cortical development at multiple ages, they predicted missing data independently with each available scan, yielding inconsistent predictions for a target time point given multiple inputs, which ignores longitudinal dependencies and introduces ambiguity in practical applications. To this end, we emphasize temporal consistency and develop a novel, flexible framework named longitudinally consistent triplet disentanglement autoencoder to predict an individualized longitudinal cortical developmental trajectory based on each available input by encouraging the similarity among trajectories with a dynamic time-warping loss. Specifically, to achieve individualized prediction, we employ a surfaced-based autoencoder, which decomposes the encoded latent features into identity-related and age-related features with an age estimation task and identity similarity loss as supervisions. These identity-related features are further combined with age conditions in the latent space to generate longitudinal developmental trajectories with the decoder. Experiments on predicting longitudinal infant cortical property maps validate the superior longitudinal consistency and exactness of our results compared to baselines’.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/3725_paper.pdf
SharedIt Link: pending
SpringerLink (DOI): pending
Supplementary Material: N/A
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Yua_Longitudinally_MICCAI2024,
author = { Yuan, Xinrui and Cheng, Jiale and Hu, Dan and Wu, Zhengwang and Wang, Li and Lin, Weili and Li, Gang},
title = { { Longitudinally Consistent Individualized Prediction of Infant Cortical Morphological Development } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15005},
month = {October},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
Obtaining a full trajectory of neurodevelopment from existing incomplete longitudinal data could improve the understanding of normal early brain development and identify neurodevelopment disorders. However, the application of these studies in real clinical settings is challenging due to limitations in previous works, such as ignoring the timing of inputs and outputs and longitudinal dependencies. To tackle these issues, the authors emphasized temporal consistency and developed a novel, flexible framework named longitudinal-consistent triplet disentanglement autoencoder to predict an individualized longitudinal cortical property development trajectory based on each available input by encouraging the similarity between trajectories with a dynamic time-warping loss.
- 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 introduce a new deep-learning method for predicting personalized longitudinal cortical property maps within 24 months of age. They utilize a disentangling strategy for interpretability, ensure longitudinal consistency for effectiveness, and employ cycle consistency for added meaningfulness and reliability. This method effectively tackles the complex task of making individual predictions using incomplete longitudinal data, a challenge that current methods have not resolved.
- 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 suggest the authors clarify the distinctions between their methodology and the one in paper [23], especially since there are similarities in notation and some modules.
- The parameters used in the models should be addressed. How did you determine them? Grid search? or else?
- 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
If possible, it would be beneficial for the authors to make the source code available to the public.
- 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 authors introduce a new deep-learning method for predicting personalized longitudinal cortical property maps within 24 months of age. They utilize a disentangling strategy for interpretability, ensure longitudinal consistency for effectiveness, and employ cycle consistency for added meaningfulness and reliability. This method effectively tackles the complex task of making individual predictions using incomplete longitudinal data, a challenge that current methods have not resolved.
- Overall, the paper is well-structured; however, I noticed some similarities with the content in paper [23]. It would be necessary for the authors to explain the distinctions between their work and the methodology used in [23].
- The parameters used in the models should be addressed. How did you determine them? Grid search? or else?
- 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?
Overall, the paper is well-structured; however, I noticed some similarities with the content in paper [23]. It would be necessary for the authors to explain the distinctions between their work and the methodology used in [23].
- 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
Accept — should be accepted, independent of rebuttal (5)
- [Post rebuttal] Please justify your decision
The authors have successfully addressed most of my concerns. Thus, it can be accepted without hesitation.
Review #2
- Please describe the contribution of the paper
This paper introduces a novel framework for predicting individualized infant cortical morphological development. Distinguished from related works, it maintains longitudinal consistency in predictions. To achieve this, the framework incorporates a novel dynamic time warping loss, ensuring similarity in predicted development trajectories. Extensive experiments conducted on validation sets validate the effectiveness of their contributions.
- 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) Significance: While numerous methods exist for predicting infant cortical morphological development, none ensure longitudinal consistency. This consistency is crucial for ensuring the meaningfulness and reliability of the results generated. 2) Novelty: The approach of employing dynamic time warping loss is both intuitive and effective, directly impacting the generated cortical property maps. 3) Extendability: This method holds potential for extension to various related tasks. While not explored in this work, it should be feasible to maintain longitudinal consistency in tasks such as functional/structural connectivity prediction and data harmonization.
- 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) Limited discussion of longitudinal consistency: Given that ensuring longitudinal consistency is the primary challenge addressed in this work, a more thorough discussion is expected, especially leveraging insights from Figure 2 to highlight the advantages of the proposed methods. 2) Inferior results in sulcal depth prediction compared to CSUNet: It is recommended that the authors discuss the trade-off they made between accuracy and longitudinal consistency in their predictions, particularly in light of the observed differences in performance compared to CSUNet. 3) Minor point: Typographical errors, such as the incorrect scale on the x-axis for individual 1 in Figure 2, were identified.
- 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure 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
1) More details should be provided for Figure 2. For instance, there is a notable rapid decrease in cortical thickness for individual 1. Is this decrease reasonable? Additionally, the observation that prediction errors are sometimes larger than the thickness variation in the ground truth raises concerns about the fairness of evaluating the proposed method’s effectiveness in preserving longitudinal consistency. I recommend that the authors compare the thickness of brain regions with obvious thickness variations rather than relying solely on the average cortical thickness. 2) For future work, I strongly recommend further testing the effectiveness of predictions in downstream tasks. For example, using predicted cortical property maps for cognition prediction, identity classification, age prediction, etc., could provide additional validation of the clinical impact of the proposed method.
- 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?
The manuscript presents a comprehensive evaluation of its contributions, offering insightful explanations and novel methodology, notably in addressing longitudinal consistency challenges. Given its potential impact and forward-thinking recommendations for future research, including downstream task validation, I recommend accepting this paper after rebuttal.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Strong Accept — must be accepted due to excellence (6)
- [Post rebuttal] Please justify your decision
The authors have fully addressed my concerns. It can be accepted without hesitation.
Review #3
- Please describe the contribution of the paper
The paper proposes a novel deep learning method to predict individualized cortical development trajectory throughout infancy (within 24 months of age) based on incomplete data. Existing methods either require complete data or predict each time point independently, leading to inconsistency. This method addresses these issues by leveraging a surface-based autoencoder, attention-based strategy, and a dynamic warping loss to predict a full trajectory that is both accurate and temporally consistent. The authors claim this is the first flexible method that can handle irregularly distributed data and achieves superior performance compared to existing techniques.
- 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.
Preserves longitudinal consistency: The model can predict individual development trajectories that are consistent with each other and closely resemble the actual trajectories based on different starting points (ages). This is validated by DTW distance and visual inspection. Improved individual prediction accuracy: Compared to other methods, the model achieves lower Mean Absolute Error (MAE) and higher Pearson’s Correlation Coefficient (PCC) for four cortical property maps (cortical thickness, sulcal depth, surface area, and myelin). Leverages disentangled features: The model separates features related to identity (subject) and age, potentially leading to more robust and interpretable predictions. Autoencoder approach: The autoencoder structure allows for efficient learning and potentially reduces the number of training parameters compared to more complex architectures.
- 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 dataset validation: The paper mentions using a 5-fold cross-validation on the training and validation set, but it doesn’t mention how well the model performs on an unseen testing set. Comparison methods: The two compared methods (CSUNet and DITSAA) might not be the most recent or state-of-the-art approaches. Including comparisons with stronger benchmarks would strengthen the evaluation. Computational cost: The paper doesn’t discuss the computational cost of training and running the model. This could be a weakness if the model is very slow compared to alternatives.
- 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure 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
In the discussion:
- Elaborate on the potential benefits of personalized interventions guided by individualized cortical property map predictions.
- Discuss potential future directions, such as incorporating additional modalities (e.g., genetics) or extending predictions beyond 24 months.
- 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?
The paper is well written and organize and every contributions are well justified and fit naturally into the problem. And their approach is novel.
- 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
Accept — should be accepted, independent of rebuttal (5)
- [Post rebuttal] Please justify your decision
Prior to the author’s rebuttal, the paper was already strong enough for recommendation. The authors have successfully addressed most of my concerns. Therefore, I recommend acceptance of this manuscript.
Author Feedback
We thank all reviewers for comments and recognition of our contributions, including 1) “offering novel methodology in addressing longitudinal consistency challenges” (R1, R3, R4); 2) “tackles the complex task which have not resolved” (R3) and “holds potential for various extension tasks” (R1); 3) “leading to more robust and interpretable predictions” (R3, R4). R1 requires detailed discussion of longitudinal consistency. Our network ensures consistent predictions over time through a temporal consistency constraint (Fig. 2). It yields predicted trajectories closely aligned with ground truth and mirrors actual developmental patterns, unlike comparison methods, highlighting our approach’s superior effectiveness. The unique decrease in cortical thickness observed in individual 1 may be due to individual differences, and our method captures these individualized patterns, demonstrating its ability to adapt to individual variability and provide precise predictions. R1 mentions the inferior results on sulcal depth compared to CSUNet. CSUNet might slightly outperform our method on sulcal depth, but has inferior performance on cortical thickness, surface area, and myelin content. One possible reason is the trade-off among 4 cortical features introduced by averaging their temporal consistency losses, aiming to minimize variability across time points. This approach may lead to a greater emphasis on dynamic features, while placing less focus on comparably stable features like sulcal depth. Consequently, our method may underemphasize the importance of sulcal depth during training, which can be solved by adjusting weights of different features in DTW loss. Notably, our method has smaller variation with the same PCC values, indicating its greater stability compared to CSUNet. R3 concerns the difference with work [23]. Our framework distinguishes itself by generating consistent results stemming from independent inferences. Our objective is to achieve precise predictions while minimizing discrepancies across multiple longitudinal results for the same individual, a key distinction from previous approaches (including [23]), and the visualization of the predicted trajectories presents solid evidence for maintaining the longitudinal consistency in Fig. 2. Additionally, our framework simultaneously predicts multiple cortical features, leveraging their implicit mutual correlation, differing from [23] which separately predicts each cortical feature. Further, our approach employs a lightweight autoencoder with an end-to-end training strategy, contrasting to the memory-intensive stage-to-stage training strategy in [23]. R3 and R4 concern the experimental details. 1) For the parameter setting, we prioritize the magnitude and significance of different losses. Specifically, the DTW loss is thousand times of others, so we manually decreased its weight. Besides, to obtain realistic predictions by separating the age and identity features while maintaining the longitudinal consistency, we tuned the weights of losses using validation datasets based on their relative importance and contributions. 2) For the computational cost, we have adopted a lightweight autoencoder framework with reduced channel numbers to minimize parameters. Our training process utilized a 3090 RTX GPU with the batch size of 16, requiring about 5 hours for 100 epochs. R4 points out the limitation of dataset and comparison methods. We apologize for the confusion on dataset description. We excluded 30% from the dataset as the hold-out test dataset, the 5-fold cross validation was conducted using the remaining 70%. We will include other longitudinal datasets to further validate our framework. Besides, we are striving to compare state-of-the-art methods. However, most works do not work directly on cortical surfaces. We will address this limitation by referencing more recent works (e.g., Ha et al., TMI, 2022 and Fawaz et al., biorxiv, 2023.10.16.562598) that can be adapted for cortical surface.
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’
As acknowledged by all reviewers, this paper introduces a novel and significant idea which was thoroughly evaluated. Minor concerns on clarity were addressed in the rebuttal.
- 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).
As acknowledged by all reviewers, this paper introduces a novel and significant idea which was thoroughly evaluated. Minor concerns on clarity were addressed in the rebuttal.
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’
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