List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Yifei Long, Jiayi Pan, Yan Xi, Jianjia Zhang, Weiwen Wu
Abstract
Low-dose digital radiography (DR) and computed tomography (CT) play a crucial role in minimizing health risks during clinical examinations and diagnoses. However, reducing the radiation dose often leads to lower signal-to-noise ratio measurements, resulting in degraded image quality. Existing supervised and self-supervised reconstruction techniques have been developed with noisy and clean image pairs or noisy and noisy image pairs, implying they cannot be adapted to single DR and CT image denoising. In this study, we introduce the Full Image-Index Remainder (FIRE) method. Our method begins by dividing the entire high-dimensional image space into multiple low-dimensional sub-image spaces using a full image-index remainder technique. By leveraging the data redundancy present within these sub-image spaces, we identify similar groups of noisy sub-images for training a self-supervised denoising network. Additionally, we establish a sub-space sampling theory specifically designed for self-supervised denoising networks. Finally, we propose a novel regularization optimization function that effectively reduces the disparity between self-supervised and supervised denoising networks, thereby enhancing denoising training. Through comprehensive quantitative and qualitative experiments conducted on both clinical low-dose CT and DR datasets, we demonstrate the remarkable effectiveness and advantages of our FIRE method compared to other state-of-the-art approaches.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43990-2_44
SharedIt: https://rdcu.be/dnwLZ
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #4
- Please describe the contribution of the paper
The main entry point of this paper is self-supervised DR/CT image denoising. A self-supervised image denoising method is designed to solve the problem of not getting noise-clean image pairs in clinical practice. In this paper, the author designs a new sub-image sampler and formulates a complete loss function for network training. Experimental results show that this method has excellent performance.
- 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 presents a sampling method based on the remainder of image index. The proposed method is convenient to understand. And the sampler design process given by the author in the paper is very detailed. Subsequently, the loss function and regularization strategy designed in this paper are highly adaptive to this sampling method. The overall logic of the article is very strong.
- 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 author gives a detailed formula in the construction part of the loss function, but the description of the formula is a little insufficient.
- 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
In this paper, the author provides detailed information such as the model and datasets. Therefore, the method is largely replicable.
- 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 formula of the loss function and regularization term in this paper can be described in more detail. The different color boxes in (b) of the first picture, such as blue and pink, should be explained by the author.
- 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?
This paper presents a sampling method based on the remainder of the image index. This method is very easy to understand, which can show the author’s good scientific research quality and active thinking. And the experimental results show that this method is very effective. In addition, it is difficult to collect noise-clean image pairs in clinical imaging. Based on practical problems, the method proposed by the author has great application prospects.
- Reviewer confidence
Very 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 proposed a novel image sampling method to complete the unsupervised denoising of noisy images. A new loss function is also proposed to correspond to the sampled image training network. Extensive experiments are also carried out to verify the proposed method.
- 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 method is novel and interesting by determining the sampled image based on the remainder of the image index. It’s more flexible with different remainders.
- The loss function designed in this paper is highly corresponding to the sampling method.
- The data sets of the experiment are diverse.
- 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 logic of the article can be more perfect. The second part of the article has less content, and it is too fragmented to separate it as a part.
- There are some grammatical mistakes, which should checked carefully.
- 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
While the authors do not provide code, they do provide detailed information about the model’s parameters, data sets, and so on. So the work seems to be largely reproducible.
- 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 related theories in the second part of the paper have less space, so I suggest the author to integrate them into other parts of the paper as a subsection.
- There are some possible grammatical problems in the article. I suggest that the author read the article carefully, and check and correct the grammar problems that may appear in the paper. For example, in the first paragraph of the introduction, ‘patients’ should be replaced with’ patients’’.
- 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?
A new self-supervised denoising method for medical images, FIRE, is proposed in this paper. This method designs a new sampler based on the remainder of the image index. Different sub-images in the sampling process can contain complete information about the original image to ensure the integrity of the original image after sampling. Experiments in this paper are relatively abundant. The article shows obvious clinical application prospects.
- Reviewer confidence
Very 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 #3
- Please describe the contribution of the paper
-
In this paper, a self-supervised technique based on the remainder of the fully image index is proposed and applied to the denoising task of medical images. In this paper, the author designs a new sampler, which realizes the subspace sampling of noise images. Moreover, this method has the advantage that sampling will not lose information.
-
This paper establishes the subspace sampling theory and explains the process of subspace sampling. Furthermore, the author explains and proves the loss combination rule of the sub-image group in detail.
-
In this paper, a new self-supervised loss function is constructed to train unsupervised image denoising networks. Regularization strategy is added to the constructed self-supervised loss to limit the subspace image training.
-
- 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 method proposed in this paper only requires the training of noisy images, which greatly reduces the difficulty of denoising clinical medical images. This paper mainly focuses on the denoising of clinical DR and CT images. Among them, the task of denoising DR images is relatively novel and has greater clinical significance.
-
The sampling method based on the remainder of the image index proposed in this paper is very interesting. Because this method is easy to understand, implementing it can reflect the author’s broadness of thought. In this paper, the author establishes the subspace sampling theory and proves that the proposed subspace sampling strategy is effective. The authors prove the proposed new self-supervised loss function and the regularization strategy. The authors also conducted related experiments to test their theory.
-
The logical and structural level of the whole article is high, which reflects the author’s high scientific research literacy and writing ability. The experimental part of the paper is substantial, the dataset is diverse, and there is a certain amount of work.
-
- 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 the Framework part of the paper, the author mentions that the value of N in the experiment is 4, and the remainder of 2i+j/4 is used to select the image. In illustrating the method, the author shows that N can have other values. However, the author does not explain whether other values of N will affect the generation of sub-images. This part of the author should explain more carefully.
-
There are some grammatical mistakes in the paper, which the author should check and correct. There are certain sentences in the paper that the author could have made more readable and enjoyable. For example, in the third paragraph of the Introduction, “parameters” should be modified to “parameter” and “introduce” to “introduced”. In the second paragraph of the Framework, “can divided” should be changed to “can be divided”.
-
The authors compare the FIRE approach to other advanced deep learning-based approaches, but few non-deep learning-based approaches are compared. This limits the ability to assess the novelty and effectiveness of the proposed approach compared to other established techniques.
-
- 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
This paper’s methodology appears to be reproducible. While the authors did not include the code, they provided comprehensive details on the model, dataset, and evaluation. Therefore, the work seems to be replicable to a significant extent.
- 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 more comprehensive literature review and grammar editing will enhance the quality of the manuscript. Further analysis of the results will also improve the paper’s quality. This paper has a good idea and has improved on previous work. It could further refine the theoretical approach and experiment on more datasets to demonstrate its full potential.
- 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?
In this paper, a new sampling method of fully image index remainder is proposed and applied to the self-supervised image denoising task. A new sampling method, loss function, and regularization term are proposed to improve the performance of a self-supervised image denoising network. The experiment was well designed, using multiple public and non-public clinical data, and was compared with a variety of other methods. Experimental results show that FIRE achieves better performance than the comparison method in terms of quantitative and visual evaluation indexes. Overall, the paper seems to be well written and presents a denoising method for DR and CT images with practical application prospects.
- Reviewer confidence
Very 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.
This paper presents a novel approach to self-supervised image denoising, leveraging a newly developed subsampling technique. The key concept revolves around utilizing subspace sampling theory to generate downsampled images, subsequently employed in training a denoising network. All reviewers unanimously acknowledged the intriguing and innovative nature of the sampling method, based on the remainder of the image index. Furthermore, the reviewers agreed that the paper is well-written, accompanied by comprehensive experimental results that effectively demonstrate the practical utility of the proposed method. Consequently, all reviewers highly recommend the acceptance of this paper.
Author Feedback
R1 Q1. The logic of the article can be more perfect. The second part of the article has less content, and it is too fragmented to separate it as a part. A1: The second part of the article is the starting point of our article. This part of the study was done by the researchers before. So we chose to use a separate chapter to introduce it and introduce the research we conducted. We’ll consider incorporating it into other parts of the paper. Q2. There are some grammatical mistakes, which should checked carefully. A2: Thank you for your careful reading of the article. We rechecked the article for possible grammatical and handwriting errors and corrected them.
R2 Q1. In the Framework part of the paper, the author mentions that the value of N in the experiment is 4, and the remainder of 2i+j/4 is used to select the image. In illustrating the method, the author shows that N can have other values. However, the author does not explain whether other values of N will affect the generation of sub-images. This part of the author should explain more carefully. A1: In this paper, we only choose the case where N equals 4 to study. But the flexibility of our method is that we can have more values of N as a remainder, and more ways to calculate the remainder. This part needs to be studied in more experiments. Q2. There are some grammatical mistakes in the paper, which the author should check and correct. There are certain sentences in the paper that the author could have made more readable and enjoyable. For example, in the third paragraph of the Introduction, “parameters” should be modified to “parameter” and “introduce” to “introduced”. In the second paragraph of the Framework, “can divided” should be changed to “can be divided”. A2: Thank you for your careful reading of the article. We rechecked the article for possible grammatical and handwriting errors and corrected them.
R4 Q1. The formula of the loss function and regularization term in this paper can be described in more detail. The different color boxes in (b) of the first picture, such as blue and pink, should be explained by the author. A1: In the calculation of the loss function, we give a detailed description. In addition, the different color boxes in Figure 1 (b), including blue and pink, respectively represent the transformation from one part of the image to this part after sampling.