List of Papers Browse by Subject Areas Author List
Abstract
Alzheimer’s Disease (AD) is a neurodegenerative condition characterized by diverse progression rates among individuals, with changes in cortical thickness (CTh) closely linked to its progression. Accurately forecasting CTh trajectories can significantly enhance early diagnosis and intervention strategies, providing timely care. However, the longitudinal data essential for these studies often suffer from temporal sparsity and incompleteness, presenting substantial challenges in modeling the disease’s progression accurately. Existing methods are limited, focusing primarily on datasets without missing entries or requiring predefined assumptions about CTh progression. To overcome these obstacles, we propose a conditional score-based diffusion model specifically designed to generate CTh trajectories with the given baseline information, such as age, sex, and initial diagnosis. Our conditional diffusion model utilizes all available data during the training phase to make predictions based solely on baseline information during inference without needing prior history about CTh progression. The prediction accuracy of the proposed CTh prediction pipeline using a conditional score-based model was compared for sub-groups consisting of cognitively normal, mild cognitive impairment, and AD subjects. The Bland-Altman analysis shows our diffusion-based prediction model has a near-zero bias with narrow 95% confidential interval compared to the ground-truth CTh in 6-36 months. In addition, our conditional diffusion model has a stochastic generative nature, therefore, we demonstrated an uncertainty analysis of patient-specific CTh prediction through multiple realizations. Our code is available at https://github.com/siyeopyoon/Diffusion-Cortical-Thickness-Trajectory.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2542_paper.pdf
SharedIt Link: pending
SpringerLink (DOI): pending
Supplementary Material: N/A
Link to the Code Repository
https://github.com/siyeopyoon/Diffusion-Cortical-Thickness-Trajectory
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Xia_Conditional_MICCAI2024,
author = { Xiao, Qing and Yoon, Siyeop and Ren, Hui and Tivnan, Matthew and Sun, Lichao and Li, Quanzheng and Liu, Tianming and Zhang, Yu and Li, Xiang},
title = { { Conditional Score-Based Diffusion Model for Cortical Thickness Trajectory Prediction } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15002},
month = {October},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper presents a novel conditional diffusion model to predict cortical thickness trajectories with the given baseline information. This has the potential to improve early diagnosis in Alzheimers disease.
- 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 presents a novel diffusion model that utilizes a 1-D attention U-Net to forecast CTh trajectories using baseline biomarker data. The approach aims to overcome the difficulty of accurately modelling disease progression with longitudinal data that are sparse and incomplete temporally. The statistical analysis demonstrates that this method outperforms existing techniques such as multi-task regression, LSTM, and RNN in predicting CTh trajectories.
- 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 clarity on the total number of biomarkers used in the proposed diffusion model makes it difficult to understand the input dimensions and the mapping process to the tensor size of 68. The baseline information provided in Figure 1 appears to come from only three categories, yet the dataset contains more categories, which raises questions about the selection process and its implications for model performance. The absence of a clear explanation regarding how the baseline information is converted into a tensor with a size of 68 creates ambiguity in understanding the data preprocessing steps. Table 2 is missing essential information about the dimensions of the U-Net architectures, particularly for “U-Net (w/o a)” and “U-Net (w/ a)”, which impedes reproducibility and comparison with other methods. If the U-Net architectures mentioned in Table 2 are 1D networks, it is unclear whether the input tensor size aligns with the 68 biomarkers used in the diffusion model, raising concerns about feature representation and model compatibility. If the U-Net architectures are 2D networks, there is no indication of what the input dimensions would be, leaving readers guessing about the nature of the input data and its suitability for the proposed models. Overall, the lack of clarity and completeness in describing the input data, preprocessing steps, and network architectures hinders the reproducibility and interpretability of the proposed diffusion model and its associated U-Net architectures.
- 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?
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
This paper presents a novel conditional diffusion model to predict cortical thickness trajectories with the given baseline information. This has the potential to improve early diagnosis in Alzheimers disease. The approach aims to overcome the difficulty of accurately modelling disease progression with longitudinal data that are sparse and incomplete temporally. The statistical analysis demonstrates that this method outperforms existing techniques such as multi-task regression, LSTM, and RNN in predicting CTh trajectories. However, the lack of clarity and completeness in describing the input data, preprocessing steps, and network architectures hinders the reproducibility and interpretability of the proposed diffusion model.
- 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?
The paper presents a novel diffusion model but details are insufficient for reproducibility.
- 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
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #2
- Please describe the contribution of the paper
- Use of a conditional diffusion-model (score based) synthesis/prediction for medical tabular data, here cortical thickness, that is able to handle missing data
- Evaluation on TADPOLE challenge with signifcant data (898 subjects with 5 distinct timepoints, with missing 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 proposed (simplified) EDM prediction model is able to handle missing data in the prediction, allows quasi-continuous prediction via a flexible selection of the time difference
- 1D Attention UNet based (across 68 regions)
- appropriate evaluation (though with limited comparison) and ablation study, showing overall good performance of the proposed method
- 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.
- In effect, quite straightforward application of a conditional diffusion model to simple tabular data, no imaging involved in the synthesis process
- Quite limited methodological novelty
- No spatial consistency in the prediction
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Do you have any additional comments regarding the paper’s reproducibility?
- The method is sufficiently straightforward that is should be reproducible based on the text alone (of course, access to code is better)
- 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
- Overall well written paper with limited need for clarification
- a larger set of comparative methods should be evaluated
- 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?
The main weakness is the lack of innovation and use of the spatial information inherent in 3D medical image data. By modeling tabular measurements (here cortical thickness, but any derived measure, like volume, surface area, etc is equally applicable) without taking account their relative spatial information, the method is taking a step back.
The proposed method is sound and has some limited innovation, with an appropriate evaluation, yet is not well suited for MICCAI given that this is really not a medical image computing paper. The input be replaced with any other measurement unrelated to imaging or medical application.
- 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
Weak Reject — could be rejected, dependent on rebuttal (3)
- [Post rebuttal] Please justify your decision
My main critique (lack of use of the spatial information inherent in 3D medical image data, i.e. method uses simply tabular data) remains. I disagree with the rebuttal that the presented work indicates this can simply be extended to 2D or 3D, or possibly a graph-based approach. Rather such extension would be non-trivial, but also would be highly interesting to the MICAI community. In its present form, I cannot recommend more than a Weak reject assessment.
Review #3
- Please describe the contribution of the paper
Paper looking at cortical thickness in the clinical course of Alzheimer’s disease. They propose a conditional score-based diffusion model specifically designed to generate cortical thickness trajectories with using clinical information (age, sex, initial diagnosis) 898 participants across five distinct time points: baseline (bl), 6 months (m06), 12 months (m12), 24 months (m24), and 36 months (m36) 178 participants (100 MCI, 78 Normal) with complete data to the testing set and the remaining 720 participants to the training sets Compared their models to other predictive models - U-net models with and without the attention mechanism (Zhou, Che, Jung) They run the model, and have a Bland-Altman analysis showing their prediction model has a near-zero bias with narrow 95% confidential interval compared to the ground-truth CTh in 6-36 months.
- 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.
Interesting paper, with a large database. Has a lot of clinical potential.
- 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.
Of the 100 MCI, after 3 years, 40 evolved to Alzheimer, 68 continued with MCI (and 8 subjects who were qualified as normal in the first scan ended up being MCI or AD). It would have been nice to describe in the methods the clinical changes of the subjects overtime, so we could understand what happened to the normal subjects that changed categories. That would also have helped to understand whether they should have been kept in the analysis or not, or if they had a specific trajectory.
- 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
Interesting and well written paper with potential clinical impact.
- 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?
Well written, large database and positive 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 provided some details
Author Feedback
Dear Chairs and Reviewers,
Thank you for your valuable feedback on our manuscript, “Conditional Score-Based Diffusion Model for Cortical Thickness Trajectory Prediction.” We appreciate the opportunity to clarify the points raised and enhance the understanding of our research contributions. We have addressed each of the main concerns as follows:
Clarity of Data and Model Description Response: 1) The demographics in Figure 1 include both age and gender variables. 2) Number of Biomarkers and Network Inputs Shape: We used 71 biomarkers: 68 cortical thickness (CTh) values for 68 brain regions, plus three non-CTh variables (i.e., age, gender, and diagnosis). These non-CTh variables, along with the time gap between scans, were vectorized by repeating each of them 68 times to match the CTh length and concatenated in the channel direction. The biomarkers and time gap were zero-padded to a dimension of 72 for model compatibility with the model’s upsampling and downsampling layers. 3) U-net Architecture: Our U-net is a 1D network. For inference, our model took 6-channel inputs (Gaussian noise, initial CTh, age, gender, diagnosis, and time gap) of length 72. The model generated the change of CTh, outputting a single-channel size of 72 through the reverse diffusion process. We then discarded the zero-padded region to obtain the final result. 4) Inclusion of Converters: As Reviewer #4 noted, our sample includes some converters (individuals whose status changed during follow-up). Including these samples allows us to provide early warnings based on MRI data changes, potentially leading to earlier intervention. Research shows that in Alzheimer’s Disease (AD) progression, structural changes often precede clinical symptoms. For example, cognitively normal (CN) subjects at baseline who later transition to mild cognitive impairment (MCI) or AD (CN converters) have smaller baseline CTh compared to stable CN individuals. We plan to further explore early-stage conversion prediction with CTh features.
Open Access to Source Code and Data Response: The data used in this study, TADPOLE, is publicly available and can be accessed through the TADPOLE Challenge (https://tadpole.grand-challenge.org/) or ADNI (https://adni.loni.usc.edu/). The source code will be made publicly available after the MICCAI review period.
Concerns about the Lack of 3D Images Utilization Response: 1) Our study focuses on CTh from T1-weighted MR images, intrinsically linked to medical image data. Our 1D U-net for estimating the score function of 1D tabular data suggests our approach can extend to 2D or 3D images. 2) The 1D CTh measurements provide quantitative indicators for brain subregions and their atrophy. Studying the temporal progression of cortical thinning may better characterize disease progression and patient condition. These measurements offer more direct interpretability than voxel-wise values in 3D imaging data. Our strategy uses a diffusion model for longitudinal, region-wise prediction tasks. For 3D images, this involves generating whole images through voxel-wise prediction, which poses computational, evaluation, and interpretation challenges. 3) We recognize that converting images to a 1D vector will lose structural priors among regions. In our model, each element of the 1D CTh vector corresponding to a brain cortex sub-region, while the diffusion model handles data distribution through the score function. We will explore extending to a graph-based diffusion model to incorporate inter-regional structural connectivity from DTI/DWI data into the prediction modeling.
Thank you again for your consideration and the opportunity to improve our manuscript.
Sincerely,
All co-authors
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’
borderline acceptance , as some reviewers comments were not clearly covered.
- 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).
borderline acceptance , as some reviewers comments were not clearly covered.
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