List of Papers Browse by Subject Areas Author List
Abstract
Glaucoma is one of the leading causes of irreversible blindness worldwide. Predicting the future status of glaucoma is essential for early detection and timely intervention of potential patients and avoiding the outcome of blindness. Based on historical fundus images from patients, existing glaucoma forecast methods directly predict the probability of developing glaucoma in the future. In this paper, we propose a novel glaucoma forecast method called Coarse-to-Fine Latent Diffusion Model (C2F-LDM) to generatively predict the possible features at any future time point in the latent space based on sequential fundus images. After obtaining the predicted features, we can detect the probability of developing glaucoma and reconstruct future fundus images for visualization. Since all fundus images in the sequence are sampled at irregular time points, we propose a time-adaptive sequence encoder that encodes the sequential fundus images with their irregular time intervals as the historical condition to guide the latent diffusion model, making the model capable of capturing the status changes of glaucoma over time. Furthermore, a coarse-to-fine diffusion strategy improves the quality of the predicted features. We verify C2F-LDM on the public glaucoma forecast dataset SIGF. C2F-LDM presents better quantitative results than other state-of-the-art forecast methods and provides visual results for qualitative evaluations.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/0498_paper.pdf
SharedIt Link: pending
SpringerLink (DOI): pending
Supplementary Material: N/A
Link to the Code Repository
https://github.com/ZhangYH0502/C2F-LDM
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Zha_CoarsetoFine_MICCAI2024,
author = { Zhang, Yuhan and Huang, Kun and Yang, Xikai and Ma, Xiao and Wu, Jian and Wang, Ningli and Wang, Xi and Heng, Pheng-Ann},
title = { { Coarse-to-Fine Latent Diffusion Model for Glaucoma Forecast on Sequential Fundus Images } },
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
This paper proposes a coarse-to-fine diffusion strategy which improves the quality of generative features for Glaucoma forecast. Several new technologies are used such as diffusion model and time sequence encoder for time-series learning.
- 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 integrates time-series image data with a diffusion model for the task of glaucoma prediction. The model is intricately designed to integrate with the data structure. Furthermore, the concurrent generation of images and classification predictions enhances the interpretability of the model.
- 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 training of the model consists of multiple phases, and the code is provided to help understand the details of the training.
- 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 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?
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
NA
- 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 network design, comprehensive experiments, well-organized paper
- 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 #2
- Please describe the contribution of the paper
This paper introduces a cutting-edge glaucoma forecasting method named Coarse-to-Fine Latent Diffusion Model (C2F-LDM), which relies on sequential fundus images. Its key contributions are two-fold: it predicts future features in the latent space to ascertain glaucoma probability and reconstructs future fundus images. Furthermore, it incorporates a time-adaptive sequence encoder that adeptly handles irregular time intervals, enabling flexible predictions. Additionally, the coarse-to-fine diffusion strategy employed enhances feature quality, resulting in superior performance compared to existing state-of-the-art methods. The interpretability of this method is emphasized through visual evaluations, underscoring its potential for clinical application in predicting glaucoma progression.
- 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）This paper introduces an innovative method known as the Coarse-to-Fine Latent Diffusion Model (C2F-LDM) specifically tailored for glaucoma forecasting. Distinguishing itself from traditional approaches, C2F-LDM offers a novel perspective by predicting future features within a latent space. 2）To enhance the quality of predictions, the model employs a sophisticated coarse-to-fine diffusion strategy.
- 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.
Although the paper compares C2F-LDM with existing methods, the comparison may be limited in scope or depth. Clinical feasibility and applicability could be further demonstrated through real-world studies involving ophthalmologists and patients. In particular, the simulation of glaucoma in the development process should be compared with real data.
- 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?
No, the paper has detailed information and resources.
- 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
There are some interesting ideas in this paper, but I still have some concerns: 1) In page 4, the structure of CLD and FLD designed in this paper is similar. Please explain in detail the difference between CLD and FLD. Why they are designed this way? 2) The authors mentioned that C2F-LDM can predict the glaucoma status at any future time point. Although the paper compares C2F-LDM with existing methods, the comparison may be limited in scope or depth. In this work, experiments predicting glaucoma at future time points are not supported by actual clinical data.
- 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 paper presents a novel approach, C2F-LDM, for glaucoma forecast using latent diffusion models, addressing an important clinical need.
- 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
The paper proposes a novel method called Coarse-to-Fine Latent Diffusion Model (C2F-LDM) to predict the future development of glaucoma. C2F-LDM utilizes historical fundus images to predict features in the latent space, which are then used to assess the likelihood of glaucoma and generate images of future fundus. The model incorporates a time-adaptive encoder to handle unequally spaced time intervals between images and a coarse-to-fine strategy to improve prediction accuracy. The authors achieved better performance than existing methods on a publicly available glaucoma forecast dataset.
- 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.
Novelty: The paper proposes a new approach (C2F-LDM) for glaucoma forecast based on latent diffusion models. This approach seems innovative and might outperform existing methods. Strong performance: The paper shows compelling quantitative results with better accuracy, sensitivity, specificity and AUC compared to other methods on the SIGF dataset. Interpretability: The framework allows for visualization of the predicted features, making the results interpretable for clinicians. Flexibility: C2F-LDM can handle irregular time intervals between fundus images and predict glaucoma status at any future time point. Potential for extension: The approach has the potential to be applied to forecast other diseases using sequential medical images.
- 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 qualitative evaluation: While the paper mentions qualitative evaluation via visualization, it could be strengthened by including more specific examples and comparisons with other methods. Lack of ablation details: The ablation study could be explained in more detail. It would be helpful to see how each component (CLD, FLD, HKE, TASE) affects the model performance individually. Missing information about limitations: The paper doesn’t mention any limitations of the proposed C2F-LDM approach.
- 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 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
Briefly discuss the limitations of the C2F-LDM, such as computational cost or generalizability to unseen data. Consider adding a short introduction to latent diffusion models for a broader audience. Briefly mention potential applications of C2F-LDM beyond glaucoma forecast.
- 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 paper proposes a novel method, C2F-LDM, for glaucoma forecast using a coarse-to-fine latent diffusion model. The paper is well-written and describes the methodology clearly. The results are promising, achieving better performance than other state-of-the-art methods.
- 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
Author Feedback
- (Reviewer #4: Q1). CLD and FLD have the same model architecture, namely conditional U-Net. The only difference between them is the conditions for guiding the latent diffusion process. For CLD, the condition is the historical knowledge encoding from historical sequential fundus images. For FLD, we introduce the results of CLD as additional condition to improve the quality of predicted features. We also add the features of the last time point from the historical sequential fundus images to provide the texture information and context constraint.
- (Reviewer #4: Q2). Thanks so much for your suggestion. We fully agree that the comparison was limited in scope or depth. We did not provide the experimental results to support our claim on predicting the glaucoma status at any future time point. The main reason is that sequential glaucoma datasets are very rare and hard to collect. In the future, we will continue to validate the effectiveness of our proposed method on more forecast tasks of other sequential medical images.
- (Reviewer #5: Q1). Our C2F-LDM still has some limitations. Firstly, C2F-LDM cannot be trained end-to-end and multi-stage training is necessary for better performance. Secondly, C2F-LDM is inability to generalize to unseen data due to the character of latent model.
Meta-Review
Meta-review not available, early accepted paper.