List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Jinpeng Li, Hanqun Cao, Jiaze Wang, Furui Liu, Qi Dou, Guangyong Chen, Pheng-Ann Heng
Abstract
In medical image analysis, anomaly detection in weakly supervised settings has gained significant interest due to the high cost associated with expert-annotated pixel-wise labeling. Current methods primarily rely on auto-encoders and flow-based healthy image reconstruction to detect anomalies. However, these methods have limitations in terms of high-fidelity generation and suffer from complicated training processes and low-quality reconstructions. Recent studies have shown promising results with diffusion models in image generation. However, their practical value in medical scenarios is restricted due to their weak detail-retaining ability and low inference speed. To address these limitations, we propose a fast non-Markovian diffusion model (FNDM) with hybrid-condition guidance to detect high-precision anomalies in the brain MR images. A non-Markovian diffusion process is designed to enable the efficient transfer of anatomical information across diffusion steps. Additionally, we introduce new hybrid pixel-wise conditions as more substantial guidance on hidden states, which enables the model to concentrate more efficiently on the anomaly regions. Furthermore, to reduce computational burden during clinical applications, we have accelerated the encoding and sampling procedures in our FNDM using multi-step ODE solvers. As a result, our proposed FNDM method outperforms the previous state-of-the-art diffusion model, achieving a 9.56\% and 19.98\% improvement in Dice scores on the BRATS 2020 and ISLES datasets, respectively, while requiring only six times less computational cost.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43904-9_56
SharedIt: https://rdcu.be/dnwH2
Link to the code repository
N/A
Link to the dataset(s)
https://www.kaggle.com/datasets/awsaf49/brats2020-training-data
https://isles22.grand-challenge.org/
Reviews
Review #2
- Please describe the contribution of the paper
The authors propose a non-Markovian diffusion model for anomaly detection in medical images, which can be used to alleviate the problem of progressive anatomical information loss in Markovian processes. On the other hand, the authors provide more guidance for the diffusion process, including input states, coarse segmentation maps, and gradients of classifiers, which allows the model to focus on anomalous regions more effectively. Finally, the authors propose a multi-step ODE solver to speed up the encoding and sampling of the model, thus reducing the computational cost for clinical applications.
- 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.
- compared to the baseline, the proposed method demonstrates the performance of sota; in addition Fig4 shows that the model constructed by non-Markovian process has better ability to retain information, and the dice coefficients of the model have consistently high scores at different timestep.
- The authors give an optimization objective for the diffusion model of non-Markovian process and derive the procedure of variational inference.
- the authors propose an ODE solver for fast encoding and sampling, and give a corresponding proof.
- 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 authors do not show the performance improvement by using the proposed ODE solver, but only mention “6 times faster” in the conclusion, which is the main shortcoming of the paper.
- The description of the graphs, formulas and their symbols are not clear enough.
- 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
Good reproducibility.
- 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 authors should describe each symbol as it appears, such as $x_{ori}, x’{ori} , \tilde{x}_t , x{seg}$ in Fig. 2, and why the three data are obtained after pre-processing?
-
The authors show the existing method of memory bank in the conditioning section of fig2, but do not give a proper description in the text.
-
the authors should explain how the gradient of the classifier is added to the sampling process, and also, why the gradient of the classifier is used instead of the output? On what labels provided by the authors did the classifier calculate the gradient? Note that the authors mention in the experimental data section that the test only contains tumor data, so is it possible that the classifier requires multiple classes (tumor or not) of data in the pre-training, and the tumor label is given directly in the test to calculate the gradient and integrate the gradient information with other condition into the noise prediction network?
Is it possible to obtain a reasonable output if tumor-free data is input at test time, i.e., the model does not detect anomalies with bias at this time. The authors should demonstrate this in their experiments.
-
This paper starts from a diffusion model for non-Markovian processes, which aims to prevent the problem of information loss in Markovian processes, but seems to degenerate into a noisy prediction task with prompt in eq.4. Although the authors give the corresponding derivation in the supplementary, it is still confusing and I hope the authors can explain this because it loses the characteristics of the non-Markovian process.
-
The “Detailed training and inferencing algorithms” mentioned in Section 2.2 of the paper are not included in the supplementary.
-
- 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?
The proposed non-Markovian process modeling idea and the ODE solution acceleration are two innovative aspects, but I feel that the authors have not reflected these two well.
- 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
Review #3
- Please describe the contribution of the paper
The paper describes a novel non-Markovian diffusion model for anomaly detection in Brain MRI, including a weakly supervised hybrid conditioning stage that enhances structural recovery and reduces noise in the obtained lesion segmentations.
- 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 focuses on weakly supervised detection, which is a very interesting and needed approach within medical imaging fields, due to the scarcity and cost of expert-annotated data.
- The results are promising for future clinical impact.
- 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 writing of the paper needs some revision to meet MICCAI standards.
- Please rate the clarity and organization of this paper
Satisfactory
- 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 used two public datasets and included references for both. Regarding the code, the authors will provide the software along with pretrained models.
- 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 authors should provide more details on the threshold filtering used to obtain the final segmentation.
-
The authors are advised to proofread the manuscript before the final submission.
-
- 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 provides a novel approach for weakly supervised anomaly detection in medical images. The experimental validation is strong, with very promising results (both quantitatively and qualitatively). However, the writing should be thoroughly revised before a final submission.
- 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
Review #4
- Please describe the contribution of the paper
The paper proposes a fast non-Markovian diffusion model (FNDM) with hybrid pixel-wise condition guidance to detect high-precision anomalies in brain MRI in a weakly supervised way. Additionally, the authors accelerate the encoding and sampling procedures in the FNDM with multi-step ODE solvers to reduce the computation burden during clinical applications.
- 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.
- proposes a novel non-Markov diffusion method that utilizes pixel-wise strong conditions for anomaly segmentation to enhance sampling speed and generation fidelity simultaneously
- largely outperforms previous SOTA diffusion model in terms of Dice metric on two public datasets
- 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.
- no runtime or computational burden comparison
- may need comparisons to more appropriate (diffusion-based) baselines
- 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
Datasets are public and authors have mentioned in reproducibility checklist that all code and pre-trained models will be made available.
- 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
- Why are there no comparison of runtimes/computational profile to baselines? The “Fast” part of the FNDM approach was not supported by results.
- for brats dataset, what was the target mask? was it the whole tumor (all tumor classes combined into a single binary class)?
- for both BraTS and ISLES, the testing was done on slices with anomalies. It will be interesting to include some results for healthy slices
- It may be interesting to have some comparisons with this baseline: https://link.springer.com/chapter/10.1007/978-3-031-16452-1_67 as it attempted to do detection/segmentation of brain anomalies with diffusion models
- “We find that FNDM significantly outperform the existing methods…” I would suggest the authors to refrain from using the word “significantly” unless they have performed statistical tests to compare the results of proposed methods vs. baselines
- 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?
The paper presents a well-motivated and well-designed approach to anomaly detection in medical images, with some scope of improvement in the results for supporting the claims in the motivation (especially regarding computational burden/time).
- 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
Review #1
- Please describe the contribution of the paper
The paper presents a fast non-Markovian diffusion model for weakly supervised anomaly detection in brain MR images.
- 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 FNDM method is novel with hybrid-condition guidance. The encoding and sampling procedures were accelerated with a novel approach. The method was explained in detail with sets of mathematical formulations and their proofs in the supplementary file.
The method was well validated with comparisons with other methods. An ablation study was carried out to show further the 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.
Experiments to demonstrate accelerated encoding and sampling are limited. The authors need to explain the context of “6-time acceleration” that they mentioned.
- 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 paper provides enough detail to reproduce its results. However, because the code is not shared, implementation details may not be repeated. Overall, the paper shows relatively high reproducibility.
- 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
A discussion of limitation and future work will be welcomed. What clinical application can it be used for? Can the method be extended to other modalities or other parts of the body?
- 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
7
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper presents a novel method with detailed explanation and sufficient validation. It is also well written and organized. I only see some minor weaknesses.
- Reviewer confidence
Somewhat 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
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.
The paper received the comments from four reviewers and all reviewers gave the positive rates. For the final version, the authors should carefully proofread the manuscript and revise the paper based on the comments raised by the reviewers.
Author Feedback
We sincerely thank all meta-reviewers and reviewers for their constructive comments and positive options on our work. We will address the comments and polish our work in the final version.