List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Zeyu Han, Yuhan Wang, Luping Zhou, Peng Wang, Binyu Yan, Jiliu Zhou, Yan Wang, Dinggang Shen
Abstract
To obtain high-quality positron emission tomography (PET) scans while reducing radiation exposure to the human body, various approaches have been proposed to reconstruct standard-dose PET (SPET) images from low-dose PET (LPET) images. One widely adopted technique is the generative adversarial networks (GANs), yet recently, diffusion probabilistic models (DPMs) have emerged as a compelling alternative due to their improved sample quality and higher log-likelihood scores compared to GANs. Despite this, DPMs suffer from two major drawbacks in real clinical settings, i.e., the computationally expensive sampling process and the insufficient preservation of correspondence between the conditioning LPET image and the reconstructed PET (RPET) image. To address the above limitations, this paper presents a coarse-to-fine PET reconstruction framework that consists of a coarse prediction module (CPM) and an iterative refinement module (IRM). The CPM generates a coarse PET image via a deterministic process, and the IRM samples the residual iteratively. By delegating most of the computational overhead to the CPM, the overall sampling speed of our method can be significantly improved. Furthermore, two additional strategies, i.e., an auxiliary guidance strategy and a contrastive diffusion strategy, are proposed and integrated into the reconstruction process, which can enhance the correspondence between the LPET image and the RPET image, further improving clinical reliability. Extensive experiments on two human brain PET datasets demonstrate that our method outperforms the state-of-the-art PET reconstruction methods.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43999-5_23
SharedIt: https://rdcu.be/dnwwB
Link to the code repository
https://github.com/Show-han/PET-Reconstruction
Link to the dataset(s)
https://doi.org/10.5281/zenodo.6361846
Reviews
Review #4
- Please describe the contribution of the paper
The paper proposes a new method in coarse-to-fine PET reconstruction from low-dose PET which utilizes two modules namely: deterministic coarse prediction module (CPM) and iterative refinement module (IRM). CPM produces a coarse reconstruction of PET and IRM samples the residual iteratively. The main idea of the paper is to delegate much of the computation to the CPM module and sample the residuals using the IRM which uses a likelihood-based model called diffusion probabilistic model (DPM). Another strategy added to the whole method is the use of two auxiliary guidance added to the low-dose PET: neighboring axial slices (NAS) and the spectrum. Lastly, a contrastive diffusion strategy was incorporated at the output level to produce the final reconstructed image. The strategy allows better association between LPET and RPET.
- 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. The authors claim that the work is the first that applies DPMs to PET reconstruction b. The proposed method tries to solve one of the disadvantages of diffusion models which is its long sampling time c. The method employs NAS and spectrum as guidance which could improve the structural integrity of the reconstructed image (does not produce “hallucinations”) d. The model performs better than previous methods (e.g. GANs, likelihood-based methods) e. Ablation study shows which part of the model could be further improved based on its contribution on the image quality and computation time.
- 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. Likelihood-based methods may produce hallucinations, the guidance in the input may help avoid this but further studies may be done b. LPET in example is not that noisy compared to SPET c. The method relies on low count and high count paired data which could be harder to implement in clinical practice
- 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 method can be reproduced if the code is provided
- 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. Include images with larger size in the paper for easier comparison. b. Include some of the limitations encountered in the experiments.
- 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 method is novel and worth exploring further. The use of diffusion probabilistic models in PET image reconstruction could improve previous SOTA results.
- 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 authors propose a diffusion-based domain transfer approach to generate standard PET images from low-dose PET images. Their approach is based on a coarse prediction of the SPET image, which is then coupled with the output of a reverse diffusion process. This two-paths method is supposed to simplify (and thus accelerate) sampling from the diffusion model. They also present auxiliary guidance strategies to improve the diffusion model output quality and clinical value.
- 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 ideas developed in the paper are very interesting and look novel. Qualitative results are impressive
- 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.
There is a lack of clarity and/or motivation for various components (see comments) which makes it difficult to really grasp why certain approaches are even tested in the first place.
- 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
It is stated on CMT that the code and model weights will be released. Please confirm.
- 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 paper lacks a bit of clarity on: How is the diffusion model trained? The input seems to be LPET + auxiliary feature maps and the output would be the difference between what the CPM predicted and the target SPET image? Is the CPM trained before the diffusion model or does this happened in parallel (changing the diffusion model objective as the CPM trains). Given that training diffusion models is expensive, is updating its target at train time a good strategy? What is the motivation for the negative sampling (contrastive loss)? Where does it come from? Please give some more details of the reasoning that lead to these decisions, as some of them seem very new.
- 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?
see comments
- 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
This paper proposed a coarse-to-fine PET reconstruction framework, including a coarse prediction module (CPM) and an iterative refinement module (IRM). The CPM generates a coarse prediction by invoking a deterministic prediction network only once, while the IRM, which is the reverse process of the DPMs, iteratively samples the residual between this coarse prediction and the corresponding SPET image.
- 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 was well organized and the clinical background and the aim were stated clearly.
- The experiments are performed in a relatively large and balanced cohort.
- 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.
- One of the main issues addressed in this paper is the challenge of “sampling from a diffusion model is computationally expensive and time-consuming, making it inconvenient for real clinical applications.” However, the author did not conduct statistical analysis and comparison on the specific sampling time. Only the amount of model parameters and calculations is not enough, the specific timeliness is more convincing.
- The rationale for the proposed coarse prediction module (CPM) is very ambiguous. As the author said, “To accelerate the sampling speed of IRM, we manage to delegate most of the computational overhead to the CPM.” So, what is the input for CPM? I mean that IRM in this paper is the denoising process in the diffusion model, which contains T steps. And the input of IRM comes from the previous step. If the input of CPM also comes from the previous step, the noisy image of the intermediate steps is obviously difficult to predict a PET image, even for coarse prediction. I hope that the author could give more explanations of the principles.
- The auxiliary guidance strategy integrates two domain information. And the frequency domain was obtained through discrete Fourier transform. But how do you integrate information from these two domains for guidance? If it’s just a simple Fourier transform between two domains, this does not seem to be more informative than using only one domain.
Overall, I find the paper to be rather dense, and hard to follow.
- 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 general libraries used in the proposed method are mentioned.
- 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 strengths and weakness section.
- 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
3
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper attempts to address an important clinical problem, but there are doubts about the rationality of some methods. Such as the comments in the main weaknesses.
- 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 diffusion-based domain transfer approach to generate standard PET images from low-dose PET images. As R1 and R3 has pointed out, the paper needs a bit more clarity about sampling being computationally expensive, model training.
Author Feedback
N/A