List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Rongjun Ge, Yuting He, Cong Xia, Daoqiang Zhang
Abstract
Multiphase contrast-enhanced computed tomography (CECT) scan is clinically significant to demonstrate the anatomy at different phases. But such multiphase scans inherently lead to the accumulation of huge radiation dose for patients, and directly reducing the scanning dose dramatically decrease the readability of the imaging. Therefore, guided with Joint Condition, a novel Circle-Supervision based Poisson Flow Generative Model (JCCS-PFGM) is proposed to promote the progressive low-dose reconstruction for multiphase CECT. JCCS-PFGM is constituted by three special designs: 1) a progressive low-dose reconstruction mechanism to leverages the imaging consistency and radiocontrast evolution along former-latter phases, so that enormously reduces the radiation dose needs and improve the reconstruction effect, even for the latter-phase scanning with extremely low dose; 2) a circle-supervision strategy embedded in PFGM to enhance the refactoring capabilities of normalized poisson field learned from the perturbed space to the specified CT image space, so that boosts the explicit reconstruction for noise reduction; 3) a joint condition to explore correlation between former phases and current phase, so that extracts the complementary information for current noisy CECT and guides the reverse process of diffusion jointly with multiphase condition for structure maintenance. The extensive experiments tested on the clinical dataset composed of 11436 images show that our JCCS-PFGM achieves promising PSNR up to 46.3dB, SSIM up to 98.5%, and MAE down to 9.67 HU averagely on phases I, II and III, in quantitative evaluations, as well as gains high-quality readable visualizations in qualitative assessments. All of these findings reveal our method a great potential in clinical multiphase CECT imaging.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43999-5_39
SharedIt: https://rdcu.be/dnwwS
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposed a novel algorithm for low dose multiphase CECT reconstruction based on a modified Poisson flow generative model. First, a progressive low dose scan and corresponding reconstruction mechanism were presented to reduce the radiation dose in multiphase CECT while maintaining the image quality. Then, in order to improve image quality, the authors proposed circle-supervision and joint condition fusing strategies to boost the refactoring capabilities of PFGM and introduce multiphase consistency and evolution information for structure maintenance, respectively. Finally, experiments showed that the proposed JCCS-PFGM outperformed several comparison algorithms and exhibited potential in 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.
In this paper, a new progressive low dose scan and reconstruction mechanism was proposed to reduce radiation dose in multiphase CECT. The first phase was allocated the highest radiation dose to obtain a fine initial image, while the dose was gradually reduced in the second and third phases. In the meantime, to preserve the image quality in all three phases, the authors improved PFGM by integrating novel circle-supervision and joint condition fusing strategies to utilize inter-phases information. Experiments and ablation studies verified the effectiveness of the proposed methods.
- 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.
a)The details of the proposed progressive low dose reconstruction mechanism are not clear. The proposed mechanism contains three parts. However, the authors only presented the structures of the third part, while the details of the first two parts were missing. b)The details of the way to penalize deviations of two normalized Poisson field are missing, as well as the way to introduce the joint condition into PFGM. c)In this paper, the definitions of causality among multiphase are not clear. And it might induce another question, whether the importance or reconstruction difficulties are the same in the three phases. It might be more rational to allocate higher doses in important or difficult phases. d)In section 2.2, the authors declared the circle-supervision strategy was capable of promoting the explicit reconstruction for noise reduction, instead of CT-similar image generation. However, PFGM basically is a generative model, and it could not change its intrinsic generative path by adding supervision between deviations of two normalized Poisson field in a circle. More explanations of the promoting explicit reconstruction ability are needed. e)In Table 2, when the radiation dose of phase two was half of phase one, the SSIM of CLEAR and DDPNet yet increased, which is confusing. f)During the experiments, the radiation dose in all three phases was set as 30%, 15% and 5%, respectively. Totally, 50% of the normal dose was used. If the 50% dose were equally allocated among three phases, whether the comparison single phase denoising methods might outperform the proposed method in the third phase, or present similar results in all three phases. The superiority of the progressive low dose mechanism and JCCS-PFGM requires further verification. g)There are several writing mistakes and confusing sentences, such as, ‘it thus extremely urgent to …’ in section 1, ‘promotes the low-radiation risk of …’ in section 1, ‘previous phase 1 and phase 2 into …’ in section 2.3 and so on.
- 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
Lack of model details and experiment settings.
- 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 details of the proposed algorithm, as well as experiment settings, should be presented. b)Details and some analysis of the ablation study should be provided. Is the ‘Joint Condition’ module ablated separately? c)The rationality of the specially designed dose reduction strategy needs further explanation. d)Some of the statements in this paper (e.g., promoting explicit reconstruction ability) require more supportive evidence. e)It is recommended to add some analysis or discussion of the experiment results, especially the confusing ones. f)Some qualitative visualization results should be moved to the main body of the paper. g)Please bold the best results in Table 1 and 2 for clarity. h)Typos in some texts and confusing sentences should be carefully revised, e.g., ‘it thus extremely urgent to …’ in section 1, ‘promotes the low-radiation risk of …’ in section 1, ‘[yphase−I, yphase−I]’ should be ‘[yphase−II, yphase−I]’ in equation (1), ‘the previous phase I yphase−II’ should be ‘phase I yphase−I’ in section 2.3, xphase-I should be replaced by xphase-II in fig. 2(a), the method ‘PFGN’ should be ‘PFGM’ in table 1, and so on.
- 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?
This paper provides a specially designed progressive low dose reconstruction mechanism for multiphase CECT. Two strategies are proposed to preserve image quality, and experiments on simulated data present the effectiveness of the proposed method. However, some details of the proposed algorithm, as well as experiment settings, are missing. Besides, several statements about specific performance or special design require more supportive evidence and further explanations. Many obvious typos appear in this manuscript that need to be carefully revised.
- 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
5
- [Post rebuttal] Please justify your decision
Though the authors further explained some details of the proposed method, there are still some major concerns, such as the rationality of the specially designed dose reduction strategy, some confusing experiment results of the comparison methods, and most importantly, the comparison results of the equal dose distribution experiments. On the one hand, since the progressive low-dose reconstruction mechanism is one of the major contributions in this paper, it is necessary to verify the effectiveness through ablation studies. On the other hand, in table 2, when the radiation dose of phase two was half of phase one, the SSIM of CLEAR and DDPNet increased and the advantage of the proposed method decreased. It requires proofs whether the proposed method could still outperform the comparison methods if the radiation dose increased to 16.6% in phase two and three. The authors did not answer all the questions, and several concerns are not satisfied in the rebuttal.
Review #2
- Please describe the contribution of the paper
The paper proposes a new method for low-dose reconstruction of multiphase contrast-enhanced CT scans. The proposed method reduces the radiation dose needed for imaging while maintaining high-quality reconstructions using a circle-supervision strategy and joint condition exploration. Extensive experiments on a clinical dataset show promising results for clinical multiphase CECT imaging.
- 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 introduces a new sub domain of LDCT denoising, i.e., progressive denoising of multi phase CT scans.
- The method is well crafted for fusing information from multi phase CT scans.
- The proposed method is extensively tested on a large clinical dataset of 11436 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.
- The paper lacks clarity in explaining certain aspects of the proposed method. For instance, Figure 3 is not properly annotated, making it difficult for readers to understand the method fully. Similarly, in Figure 4, it is unclear how the self-attention-based module will function for phase 1 and phase 2 images. While the overall idea of the proposed method is understandable, the paper does not provide enough information to make it easily reproducible for readers.
- The comparison with other methods in the paper is not justified. As mentioned in the supplementary material, the effective radiation dose in each phase is approximately 16%. The previous method, which only used images from a single phase to denoise that specific phase’s image, would require an image with a 16% dose for all phases. However, in this study, the authors used a 5% dose in the final phase and a 30% dose in the first phase for evaluating their method. Therefore, other methods should use images with a 16% radiation dose for each phase to make a fair comparison. Moreover, the difference between different methods is not significant for a 15% dose, and recent state-of-the-art methods can even outperform the proposed method for this radiation dose.
- There is a significant concern about the clinical impact of fusing information from a multiphase CT scan. As different abnormalities, such as lesions or tumors, change in different phases, the information provided by each phase is essential for accurately diagnosing underlying conditions. If any features are compromised due to superimposition, it could result in serious complications and even misdiagnosis. One possible way to address this concern is to ask radiologists to track changes in a few abnormalities present in the scans to ensure that this issue is not occurring.
- 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
Not enough for reproduction.
- 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
See Weakness 3.
- 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?
The main reason behind the rating is the weakness 1 and 2. Weakness 3 can improve the quality of the study and make it a better paper but is not a reason for the rating at its current form.
- 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
5
- [Post rebuttal] Please justify your decision
I agree with the other reviewer’s positive evaluation of this work’s potential and relevance to the MICCAI community. However, I find the author’s response to weakness 2 less convincing. I suggest including a paragraph dedicated to discussing the clinical impact of the proposed method. Being one of the pioneering approaches in the field, it would be beneficial to highlight the significance and practical implications of this work. Additionally, providing a more detailed explanation of the method itself would enhance the reader’s understanding and comprehension, allowing them to grasp the technical aspects more effectively. These revisions would further strengthen the paper’s appeal and contribute to its value within the MICCAI community.
Review #3
- Please describe the contribution of the paper
The paper proposes JCCS-PFGM, a deep learning method for multi-phase CECT denoising. This is an interesting paper for progressive low-dose CT reconstruction for multiple phases. Their proposed method showed decent improvement over previously proposed methods, both quantitatively and qualitatively.
- 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.
A novel algorithm consisting of (1) progressive low-dose reconstruction for multi-phase CT that utilizes the consistency between different phases. This was not considered in previously-proposed methods. (2) Circle-supervisor strategy embedded in a diffusion model. (3) And a joint condition that supervise multi-phase reconstruction. The proposed method demonstrates reasonably amount of improvement over previously-proposed method, both quantitatively and qualitatively.
- 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) Lack of clinical evaluation. (2) Lack of statistical testing on quantitative measurements. (3) The proposed method consists of multiple complex deep learning modeules. I suspect the model will need much longer time for training/testing compared with other models. Especially due to the 2 diffusion/reversion processes embedded in the network. A comparison for training/testing time would be preferred. (4) Because of the 2 diffusion/reversion processes, is the proposed model significantly more memory-intensive compared to other models? (5) How did the authors evaluated the models proposed for single-phase CT? Did the authors train the network separately for different phases? Or the authors combine all the phases together for training? (6) Are the dose levels for phases 1-3 fixed? Is phase 1-3 always 30%, 15%, and 5% in clinical settings? What will happen if dose levels change during evaluation? (7) Why phase 1 have highest dose compared to other phases? Is it because of the clinical acquisition workflow? Could we have 5% dose for phase 1 and 30% dose for phase 3? How will this affect network performance? Will the network adjust for dose changes? (8) Are the reconstructions 2D? If it is 2D, how hard will it be to transfer to 3D? or at least 2.5D?
- 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
Code is available, so it should be easy to reproduce the results.
- 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
See my comments before.
- 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?
Based on my comments, this is an interesting and well-written paper for multi-phase CT denoising with some weaknesses detailed before. But I think this is a good paper for MICCAI as a conference paper, especially considered the length of the 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
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 proposes a new algorithm for low-dose reconstruction of multiphase contrast-enhanced CT scans. The proposed method reduces the radiation dose needed for imaging while maintaining high-quality reconstructions using a circle-supervision strategy and joint condition exploration. The proposed JCCS-PFGM outperformed several comparison algorithms and exhibited potential in clinical applications. The major concerns include: (1) lacks clarity in-detail explanation of the proposed method (2) Lack of clinical evaluation and impact. (3) Lack of statistical testing on quantitative measurements. (4) Comparison with other methods are not well justified.
Author Feedback
We thank R1 and R3 for accepting our paper as “good paper for MICCAI”, “novel”, “decent improvement” and “well-written”, and thank R2 for appreciating our work as “a new sub domain of LDCT denoising”, “promising results”, “well crafted”.
We thank AC for supporting our work as “high-quality”, “potential in clinical applications”, and summarizing 4 suggested rebuttal points.
1.Details of proposed method {R1&R2: progressive low dose reconstruction(PLDC) & Fig.4} The structure of PLDC is explained in Fig.2a and Eq.1, that latter phase CT is reconstructed with the priori knowledge from former phases for complementary. Following PLDC mechanism, as shown in Fig.2b and 4, for phase III, the self-attention of phases II & I and the further cross-attention with phase III are combined to guide reconstruction. By analogy, phases II combines the self-attention of its former phase I and the further cross-attention with itself; phases I only need convolution for embedding because there are no former phase. {R1} The deviations of two normalized Poisson field (NPF) is penalized by minimizing their difference to drive the accurate recovery and guide the purposive estimation of NPF.
{R2} Fig.3 is annotated with text explanation in subsection 2.2 and consistent with concept in diffusion model. Initial CECT is original CT at t0. Two Forward are perturbed procedure with same noise distribution. Two Reverse are using DNN to estimate NPF. Temporary CECT is result restored with Poisson field. Secondary perturbed and image and NPF are get based on temporary CECT.2.{R2,R3}Clinical impact 1)As a collaborative research with radiologist, the results of mutiphase reconstruction on clinical data are carefully checked by radiologist and get approved. The fusion of information from a multiphase CT do not impair the reconstruction among different phases. It is beneficial from the information selection in multiphase joint condition via cross-attention between between former and current phases. According to the inter-phase correlation, the interference information get adaptively suppressed and the correlated complementary is enhanced. 2)Our method eliminates redundancy and interference in the former phase, and deeply explores the inter-phase correlation with joint condition, so that drives the reconstruction to focus more on specific structure in each phase. Therefore, the total dose of multiphase CT is greatly reduced, and anatomical structure gets effective imaging, thank to such inter-phase prior knowledge.
3.{R3}Statistical test For CT reconstruction methodology research papers, the mean of reconstruction error used in quantitative analysis is the most widely recognized statistical parameter and primary index in the evaluation system. This is because it can objectively reflect the central tendency of reconstruction errors, which is the most important factor to effectively and intuitively evaluate and compare reconstruction performance.
4.{R1,R2}Comparison with other methods 1)The evaluation of all methods on single phase reconstruction can be seen from the results of phase I in Tab.2 and Fig.S. Our method still obviously gains best performance, even without multiphase collaboration. 2)To explain the effect of multiphase collaboration on multiphase CT reconstruction, all methods are compared with same dose distribution for fairness. The comparison aims to verify that combining the correlation between multiphases can promote reconstruction, which is ignored by existing methods. The results on phase II and III prove this superiority, and even at ultra-low dose, our method still gets the best results compared to the others.
Post-rebuttal Meta-Reviews
Meta-review # 1 (Primary)
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The innovation and clinical impact of this work using progressive approach to reconsutruct low-dose multiphase CECT images are clear. The authors have made an effort to address some of the reviewers’ concerns, but not all of the major ones. I suggest the authors include ablation studies and statistics tests (could be in the supplement) and address the unexpected performance in Table 2, in their final version.
Meta-review #2
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
I have read the comments and rebuttal. This paper is about progressive low-dose multiphase CECT reconstruction. The concerns raised by the reviewers are partially addressed in the rebuttal. The authors are suggested to take their comments into consideration if the paper is accepted.
Meta-review #3
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
This paper presents an interesting method for low dose reconstruction of multiphase CECT images. The rebuttal addresses the broader concerns highlighted by the reviewers. Statistical measures such as paired t-test etc over the reported reconstruction metrics can be reported in the final version along with the justification for the 30%, 5% doses.