List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Jiaqi Cui, Pinxian Zeng, Xinyi Zeng, Peng Wang, Xi Wu, Jiliu Zhou, Yan Wang, Dinggang Shen
Abstract
To obtain high-quality positron emission tomography (PET) images while minimizing radiation exposure, various methods have been proposed for reconstructing standard-dose PET (SPET) images from low-dose PET (LPET) sinograms directly. However, current methods often neglect boundaries dur-ing sinogram-to-image reconstruction, resulting in high-frequency distortion in the frequency domain and diminished or fuzzy edges in the reconstructed images. Furthermore, the convolutional architectures, which are commonly used, lack the ability to model long-range non-local interactions, potentially leading to inaccurate representations of global structures. To alleviate these problems, in this paper, we propose a transformer-based model that unites triple domains of sinogram, image, and frequency for direct PET reconstruc-tion, namely TriDo-Former. Specifically, the TriDo-Former consists of two cascaded networks, i.e., a sinogram enhancement transformer (SE-Former) for denoising the input LPET sinograms and a spatial-spectral reconstruction transformer (SSR-Former) for reconstructing SPET images from the denoised sinograms. Different from the vanilla transformer that splits an image into 2D patches, based specifically on the PET imaging mechanism, our SE-Former divides the sinogram into 1D projection view angles to maintain its inner-structure while denoising, preventing the noise in the sinogram from prorogating into the image domain. Moreover, to mitigate high-frequency distortion and improve reconstruction details, we integrate global frequency parsers (GFPs) into SSR-Former. The GFP serves as a learnable frequency filter that globally adjusts the frequency components in the frequency domain, enforcing the network to restore high-frequency details resembling real SPET images. Validations on a clinical dataset demonstrate that our TriDo-Former outperforms the state-of-the-art methods qualitatively and quantitatively.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43999-5_18
SharedIt: https://rdcu.be/dnwww
Link to the code repository
https://github.com/gluucose/TriDoFormer
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes a new approach called TriDo-Former for reconstructing high-quality positron emission tomography (PET) images from low-dose PET sinograms. The proposed method uses a transformer-based model that unites sinogram, image, and frequency domains to reconstruct PET images directly, which also uses global frequency parsers (GFPs) to adjust the frequency components in the frequency domain, improving the reconstruction details and restoring high-frequency details. The proposed method outperforms state-of-the-art methods both qualitatively and quantitatively on a clinical 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.
-
This paper is well-written and thoughtfully presented. The authors provide a clear and logical organization of their ideas, which enhances the paper’s readability.
-
The experiments are conducted with a high degree of rigor and are validated on both normal and abnormal populations, making the findings more compelling.
-
The proposed GFP is well-motivated and shows promising results in addressing high-frequency details, which is critical for PET reconstruction.
-
- 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.
- While the paper presents a well-executed combination of existing algorithms, it lacks novelty in terms of PET reconstruction techniques. Specifically, the proposed TriDo-Former could be applied to any modality reconstruction task, and therefore does not provide unique insights or advancements in the field of PET reconstruction.
2 Although the proposed GFP approach theoretically offers advantages in high-frequency detail processing, the experimental results do not adequately demonstrate its superiority. The quantification and visualization analyses are insufficient to support the claim of GFP’s advantages in this aspect.
- Some parts of the experimental section are not clearly explained, such as the classification method used for downstream tasks in MCI diagnosis. The authors do not present any details of the classification model.
- 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
The paper lacks open-source codes, and it appears that the data used may be private. However, since the proposed method is not overly complex, and the implementation details are provided in the paper, it is theoretically possible 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
-
The most significant innovation in the paper lies in the use of GFP and the downstream validation of MCI classification. The former represents a significant advantage and highlight of the algorithm in the reconstruction task, while the latter is a more clinically meaningful downstream task specifically designed for PET reconstruction. Therefore, I recommend that the authors strengthen the explanations of these two aspects in the experimental section. For example, they could provide visualizations of GFP and additional details regarding the MCI classification task.
-
It would be beneficial for the paper to include the computational and parameter requirements in the experimental comparison table. Clinical applications are sensitive not only to performance but also to speed and feasibility.
-
- 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 novelty of the paper may be relatively insufficient compared to existing literature. Additionally, the experimental results could be strengthened to provide more compelling evidence. The authors could consider including missing details that are necessary to replicate their experiments.
- 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
The authors partly solved my concerns. While some are not solved. I improve my rating to weak accept consequently.
Review #2
- Please describe the contribution of the paper
The authors propose an end-to-end architecture based on transformer that utilizes triple domains of sinogram, image, and frequency to directly reconstruct the clinically acceptable SPET images from LPET sinograms.
- 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 is very well-written, easy to follow and addresses an important issue. The authors describe the problem with PET images lucidly and how the proposed method alleviates it. The description of the method is very detailed.
- The results (PSNR and SSIM) look promising.
- Adequate amount of literature review, experiments have been done.
- 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 utilize transformer architecture to account for distortion of the reconstructed image in the high-frequency of the frequency domain, eg. blurred edges and to model long range semantic dependencies (such as missing or inaccurate global structure) properly. An additional figure showcasing reconstructed images with examples of the above two and comparing that with state-of-the-art is advised.
- Points 2 and 3 under Contributions (“The contributions of our proposed method can be described as follows …”) should not be mentioned under contributions since it’s a part of the architecture and directly follows the previous statement, “To our knowledge, we are the first to leverage both triple-domain knowledge and transformer for PET reconstruction.”. Please mention the strengths of the proposed method such as qualitative/quantitative performance when compared to the state-of-the-art.
- 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 seems to be reproducible since the architecture has been described in details. The authors also provide further information for reproducibility under section 2.4 Details of Implementation
- 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
My comments are very minor and mostly deals with reformatting some sentences and adding a figure or two. Please see Weakness section for details.
- 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?
This is a very well written paper with good contribution. Problem description and clinical significance of the problem is spot on. The authors did a through literature review and analysis on state-of-the-art methods in this area. The results look promising.
- 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
6
- [Post rebuttal] Please justify your decision
The authors addressed most of the comments that were made by the reviewers. So my final decision is accept
Review #3
- Please describe the contribution of the paper
The paper proposes a transformer-based low-dose PET reconstruction method that exploits triple-domain knowledge in sinogram, image, and frequency domains. The experiments demonstrate outstanding performance in terms of image accuracy and downstream disease diagnosis tasks.
- 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 transformer-based low-dose PET reconstruction method effectively utilizes triple-domain knowledge in sinogram, image, and frequency domains.
- The experiments showcase excellent performance in terms of image accuracy and downstream disease diagnosis tasks, highlighting its potential for practical applications.
- 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 regarding the evaluation of clinical diagnosis. It is not clear whether the diagnosis is conducted by radiologists or a specific AI tool.
- 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
The paper includes sufficient details to ensure 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
Please provide more details about the evaluation of clinical diagnosis to improve the clarity and understanding of the methodology.
- 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 presents a novel and effective method for PET reconstruction, with promising results in terms of image accuracy and downstream disease diagnosis tasks. However, more clarity is needed regarding the evaluation of clinical diagnosis.
- 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.
In the paper the authors present a transform-based low-dose PET reconstruction method. The paper received two accepts and a reject. I invite the authors to address the issues raised by all reviewers in the rebuttal phase.
Author Feedback
We thank all the reviewers (R1, R2, R3, Meta-R) for their constructive comments, which have been carefully addressed as follows: Q1: The method lacks novelty in terms of PET reconstruction techniques, which could be applied to any modality reconstruction task. (R1) A1: We would like to highlight our novelty in terms of PET reconstruction techniques as follows: First, and most importantly, our designed SE-Former is tailored for the sinogram domain of LPET to suppress the noise, which can hardly be implemented in other modalities such as MRI due to different imaging mechanisms. Specifically, considering that each row of sinogram is the 1D projection at an imaging view angle, our SE-Former tactically splits sinogram into 1D projection view angles instead of adopting the commonly-used 2D patches as input, thus preserving the continuity of each view. Second, we leverage the TransEncoder blocks to model interrelations among different view angles, thus maintaining the sinogram inner-structure while effectively reducing the noise. Third, we propose a GFP module to calibrate the easy-to-lose high-frequency details due to intrinsic fuzzy properties of PET images. We contend that these efforts can provide some unique insights for PET reconstruction. We will clarify our novelty in the final paper. Q2: Details about clinical diagnosis. (R1&R3) A2: The classification model in the clinical diagnosis experiment is a multi-layer CNN which is first trained by real SPET images, thus enabling to distinguish between NC subjects and MCI subjects with 90% accuracy. Then, we evaluate the PET images reconstructed by different methods on the trained classification model. Our insight is that, if the model can discriminate between NC and MCI subjects from reconstructed images more accurately, the quality of the reconstructed images and the quality of SPET images (preferred in clinical diagnosis) are closer. Thus, the highest classification accuracy achieved by images reconstructed by our method can demonstrate its superiority in supporting clinical diagnosis. We will include these details in the final paper. Q3: Quantification and visualization of GFP. (R1) A3: Ablation study in Table 2 shows that our GFP improves PSNR by 0.329 and 0.396 for NC and MCI subjects, respectively, which quantitatively demonstrates its effectiveness. Note that achieving such amount of enhancement is not easy, as supported by comparing the performance gains obtained by comparison methods in Table 1. Furthermore, we compared the results of the model with and w/o GFP in the downstream clinical diagnosis experiment. The model w/o GFP achieves 86.7% classification accuracy, which is 1.9% lower than the model with GFP, verifying the crucial role of GFP in enhancing clinical significance of reconstructed images. In addition, we analyzed 2D and 1D power spectrum of the reconstructed images and found that the high frequency parts of images reconstructed by the model with GFP are more similar to those of SPET images, visually proving the contribution of GFP. These results and analysis will be included in the final paper. Q4: Computational and parameter requirements. (R1) A4: Parameters of DeepPET, Sino-cGAN, LCPR-Net, 3D-cGAN and TriDo-Former (proposed) are 60M, 39M, 77M, 127M and 38M, respectively. GFLOPs of above methods are 49.20, 19.32, 77.26, 70.38 and 16.05, respectively. Thus, our method yields the least parameter and computational requirement, demonstrating its speed and feasibility in clinical applications. These results will be included in the final paper. Q5: Reformat some sentences and add a figure or two. (R2) A5: As suggested, we will reformat these sentences and highlight our superior performance in Contributions. We will also provide figures of reconstructed images with distortions and compare them with state-of-the-art in the final paper. Q6: Reproducibility. A6: Our code will be released at https://github.com/gluucose/TriDoFormer. We will provide a link in final paper.
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 paper received two accepts and one reject in the initial review phase, and then was brought to the rebuttal phase. The authors provided a very nice rebuttal to address all reviewers’ concerns. I looked at the rebuttal as well. The authors successfully convinced R1 in their rebuttal and R1 recommended accepting the paper in the post-rebuttal evaluations. So I am happy to recommend to accept the paper for the publication of MICCAI 23.
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.
All three reviewers now have unanimously given accept for this work and the rebuttal satisfied concerns from previous reviews.
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.
The authors have sufficiently addressed the comments regarding the novelty, clinical diagnosis application, computational cost, and parameters of their model. They have noted that the reviewers’ comments will be incorporated into the final version of the paper. Hence, I suggest accepting this paper for MICCAI.