Authors

Jiacheng Wang, Jing Yang, Qichao Zhou, Liansheng Wang

Abstract

Skin lesion segmentation in dermoscopy images has seen recent success due to advancements in multi-scale boundary attention and feature-enhanced modules. However, existing methods that rely on end-to-end learning paradigms, which directly input images and output segmentation maps, often struggle with extremely hard boundaries, such as those found in lesions of particularly small or large sizes. This limitation arises because the receptive field and local context extraction capabilities of any finite model are inevitably limited, and the acquisition of additional expert-labeled data required for larger models is costly. Motivated by the impressive advances of diffusion models that regard image synthesis as a parameterized chain process, we introduce a novel approach that formulates skin lesion segmentation as a boundary evolution process to thoroughly investigate the boundary knowledge. Specifically, we propose the Medical Boundary Diffusion Model (MB-Diff), which starts with a randomly sampled Gaussian noise, and the boundary evolves within finite times to obtain a clear segmentation map. First, we propose an efficient multi-scale image guidance module to constrain the boundary evolution, which makes the evolution direction suit our desired lesions. Second, we propose an evolution uncertainty-based fusion strategy to refine the evolution results and yield more precise lesion boundaries. We evaluate the performance of our model on two popular skin lesion segmentation datasets and compare our model to the latest CNN and transformer models. Our results demonstrate that our model outperforms existing methods in all metrics and achieves superior performance on extremely challenging skin lesions. The proposed approach has the potential to significantly enhance the accuracy and reliability of skin lesion segmentation, providing critical information for diagnosis and treatment. All resources will be publicly at https://github.com/jcwang123/MBDiff.

Link to paper

DOI: https://doi.org/10.1007/978-3-031-43901-8_41

SharedIt: https://rdcu.be/dnwDP

Link to the code repository

https://github.com/jcwang123/MBDiff

Link to the dataset(s)

N/A

Reviews

Review #3

Please describe the contribution of the paper

This work introduces a novel diffusion-based approach for boundary evolution used in the application of skin lesion segmentation. The skin lesion segmentation task is modelled as a boundary evolution problem, featuring multi-scale features from pretrained models, as well as an uncertainty-based ensample at inference time. Experiments on public datasets comparing the method with other SOTA methods prove the usefulness of this approach.
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 problem formulation is novel for lesion segmentation.
- The paper is well structured, clear and easy to read.
- The baseline comparison is thorough, featuring a wide selection of relevant baselines, thereby supporting the claims of the paper.
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 improvement shown in this work is incremental.
- Due to the fact that there was no validation-set based model selection but instead the training was stopped after a fixed number of iterations, the differences observed between models could have been due to chance, especially compared to the Xbound-Former, where the differences are really minor.
- The methodological contribution is only explained in the text, making it difficult to understand. The paper could have really benefited from a figure showing an overview of the methodology.
- The metrics chosen for evaluation do not feature a separate analysis of false positive and false negatives (e.g. precision and recall)
- The ablation analysis in Figure 3 does not explicitly display the mean values, which are quite important for a comparison with the other 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

The experiments are carried out on publicly available data. There are sufficient details provided in the manuscript about the experiments, and the authors claim to release the code as well as trained models upon acceptance.
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
- Both IOU and Dice metrics convey the same information, and one can be derived from the other. I do not see the usefulness of reporting both.
- For more suggestions, please refer to the weaknesses.
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

4
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

Incremental paper with some experimental limitations. Please refer to the weaknesses portion of the review for the limitations.
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

Review #1

Please describe the contribution of the paper

This paper proposes the Medical Boundary Diffusion Model which outperforms existing methods in all metrics and achieves superior performance on extremely skin lesions. To start with, this paper introduces an efficient multiscale image guidance module that constrains the evolution of boundaries to align with the desired lesions. Additionally, this paper presents an evolution uncertainty-based fusion strategy that refines the evolution results and yields more precise lesion boundaries.
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.

This paper is well structured and clearly described so it is easy to be understood. Based on the previous groundwork, the diffusion model is used on the segmentation task of skin diseases, and the uncertainty-based boundary determination method is creatively proposed.
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) Need to compare with MedSegDiff-v2 to show this work is enough solid. (2) Need to show more expandability of this work. It means, this method seems not just to be suitable for skin diseases. So, could this method be expanded to other areas?
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

It is believed that this work can be 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

Try to find more dataset and prove that this method can be applied in other medical images segmentation task.
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 work is complete and reasonable and has some practicality, but expandability needs further exploration.
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 #2

Please describe the contribution of the paper

Authors propose MB-Diff, a new semantic segmentation architecture. MB-Diff essentially is a diffusion model to obtain a clear segmentation map through evolving within finite times. Compared to other end-to-end segmentation architectures in skin lesion segmentation, MB-Diff formulates skin lesion segmentation as a boundary evolution process to thoroughly investigate the boundary knowledge. Authors validate the method on two segmentation tasks, outperforming baseline methods from literature.
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.

Robustness for various lesion sizes: MB-Diff’s main strength is that there is no struggling with extremely hard boundaries, such as particularly small or large sizes lesion. Because the size of image in the evolution of diffusion model is not changed, which can avoid the small size lesion missing and extract features in global receptive field. Extendability: the method can be applied to already existing brain cancer segmentation problems with minor changes to the model architecture. Even though this is not investigated in this work, an extension to 3D segmentation should be straightforward. Interpretation: Authors evaluate the results’ differences using uncertainty estimation.
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 comparison to state-of-the-art: a comparison of meddiff-v2 should be added. Because of its better performance and the use of transformer structure. Limited discussion of multi-scale feature: The architecture is very similar to MedSegDiff in that the condition encoder is changed to PVT to extract multiscale features, but no corresponding ablation experiments are added and the performance of multiscale features is not discussed. No statistical evaluation of the results: paired tests would give statistical weight to the argument of superiority of the proposed method.
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

It’s 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

Lack of clarity: More information implementation: The authors do not mention in this paper the number of parameters of the model, the computational resources and training time used for model implementation, and the time used for model inference. This can be an obstacle to evaluating the usefulness of the model in practice. For future work, I would recommend: Extension to 3D: the robustness and compactness makes this approach particularity attractive for 3D segmentation. Explore performance on many more problems: MB-Diff could be universally applicable, but here it is used on only a few tasks. I would strongly recommend to apply MB-Diff to the medical image segmentation decathlon. Also, MB-Diff could be directly used on X-ray and ultrasound segmentation task.
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?

Because the proposed method significantly outperform the other studies, and it has good interpretation.
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

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.

Strength: 1) The paper is well structured and clearly described. 2) Diffusion model and uncertainty-based boundary determination are two novel technical contributions for skin lesion segmentation. 3) The proposed methods are robust to lesion size. 4) The baseline comparison is thorough, featuring a wide selection of relevant baselines, thereby supporting the claims of the paper.

Weakness: 1) The ablation analysis in Figure 3 does not explicitly display the mean values, which are quite important for a comparison with the other baselines. 2) Lack of implementation information, including the number of parameters of the model, the computational resources and training time used for model implementation, and the time used for model inference. 3) Need to compare with MedSegDiff-v2 to show this work is enough solid.

Author Feedback

N/A

back to top

Medical Boundary Diffusion Model for Skin Lesion Segmentation