List of Papers Browse by Subject Areas Author List
Abstract
The dopamine transporter (DAT) imaging such as 11C-CFT PET has shown significant superiority in diagnosing Parkinson’s Disease (PD).
However, most hospitals have no access to DAT imaging but instead turn to the commonly used 18F-FDG PET, which may not show major abnormalities of PD at visual analysis and thus hinder the performance of computer-aided diagnosis (CAD).
To tackle this challenge, we propose a Metabolism-aware Anomaly Detection (MetaAD) framework to highlight abnormal metabolism cues of PD in 18F-FDG PET scans.
MetaAD converts the input FDG image into a synthetic CFT image with healthy patterns, and then reconstructs the FDG image by a reversed modality mapping.
The visual differences between the input and reconstructed images serve as indicators of PD metabolic anomalies.
A dual-path training scheme is adopted to prompt the generators to learn an explicit normal data distribution via cyclic modality translation while enhancing their abilities to memorize healthy metabolic characteristics.
The experiments reveal that MetaAD not only achieves superior performance in visual interpretability and anomaly detection for PD diagnosis, but also shows effectiveness in assisting supervised CAD methods.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/1687_paper.pdf
SharedIt Link: https://rdcu.be/dV1Ol
SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72069-7_28
Supplementary Material: https://papers.miccai.org/miccai-2024/supp/1687_supp.pdf
Link to the Code Repository
https://github.com/MedAIerHHL/MetaAD
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Hua_MetaAD_MICCAI2024,
author = { Huang, Haolin and Shen, Zhenrong and Wang, Jing and Wang, Xinyu and Lu, Jiaying and Lin, Huamei and Ge, Jingjie and Zuo, Chuantao and Wang, Qian},
title = { { MetaAD: Metabolism-Aware Anomaly Detection for Parkinson’s Disease in 3D 18F-FDG PET } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15002},
month = {October},
page = {291 -- 301}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors introduce Metabolism-aware Anomaly Detection (MetaAD) for anomaly detection in 18F-FDG PET scans for Parkinson’s Disease (PD).
- 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.
Comprehensive analysis Multiple evaluation metrics (HMA, LMA, AMR, AMC) Well designed ablation study Utilizing the proposed unsupervised method to enhance supervised pipelines
- 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.
Single dataset, single institution Limited novelty from the ML point of view The proposed model could be compared with more modern pipelines (e.g., more advanced variations of UNET, diffusion models, etc)
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
in-house only studies can never be considered reproducible. However, given the high tempo of the research in the healthcare domain, and challenges of finding open-source datasets and open-sourcing the internal data, we need to accept this type of research.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
As a reviewer, my number one goal is to help you improve your manuscript, rather than judging if it should be published or not. I am aware of how difficult it is to prepare a manuscript at this level, and I wish you good luck with this publication. Introduction contains several modules from methods. As an example, Figure 1 should be in methods. At the end of the introduction, provide the contributions in the form of bullet points. Add a flowchart/figure in 3.1 to visualize data splitting, patient numbers, etc. Add a section/paragraph to highlight the limitations and future work.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Weak Accept — could be accepted, dependent on rebuttal (4)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The manuscript is well-written and it is based on a comprehensive analysis. Novelty is the only concern. If a reviewer with relevant clinical expertise can confirm the application is novel, the manuscript should be accepted.
- Reviewer confidence
Very confident (4)
- [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
The paper proposes a deep learning tool for unsupervised anomaly detection (UAD) in Parkinson’s disease from FDG-PET images. It is based on a cycleGAN that converts FDG-PET into CFT-PET and back to FDG.
- 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 model has a CycleGAN path but also features a novel parallel path that generates synthetic anomalies to help the model learn eliminate those anomalies.
- The experimental part is strong and convincing, the method is compared to other UAD and in combination with supervised methods. There is also a nice ablation study.
- 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 evaluation is performed in a single dataset that is not public.
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
N/A
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
The total loss in eq. 3 has four terms but only two are explicitly indicated (L_cycle and L_con), the remaining two are only mentioned. Figure 2 should include the discriminators. There is a typo in the word abnormal in the title of section 2.2 Reference 8 is incomplete, missing the publication title In the references, abbreviations and acronyms (MR, MRI, FDG, etc…) should be in capital letters, and parkinson should be changed to Parkinson.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Accept — should be accepted, independent of rebuttal (5)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The proposed method has novelty, and the experimental part is complete and convincing that it outperforms competing approaches.
- Reviewer confidence
Confident but not absolutely certain (3)
- [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
This paper demonstrates a novel idea for detecting parkinsons via FDG PET images rather than the less widely available but more effective DAT imaging. It does this by creating a synthetic DAT (healthy) image and converting that back to FDG image and determining the differences. As the FDG will have been made from a healthy DAT image the difference should indicate a metabolic abnormality.
- 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.
Genuinely clever idea. Only reminds me of a few similar papers and then even then only slightly, compared to a lot of papers in this field which are only small variations. The main novel aspect is the use of the data that faces inherent limitations such as low availability or pairing, and the FDG -> DAT -> FDG transition. Considers real clinical limitations and addresses them. Addresses why they chose their methods. Would count as explainable AI
- 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.
could possibly give more info for reproducibility
- Please rate the clarity and organization of this paper
Excellent
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
I was between the bottom two and not 100% certain enough information is available for reproducibility.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
Not necessary but I would be interested in your thoughts about why the other UAD methods in 3.2 do not perform as well. Or what element of your method makes it better? The images used for training, is there a good demographic distribution in the patients?
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Strong Accept — must be accepted due to excellence (6)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Fairly novel method and to the best of my knowledge a novel application. Addresses a realistic issue.
- Reviewer confidence
Very confident (4)
- [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
Author Feedback
Thanks for the valuable comments. We appreciate the recognition of our work and have addressed each of the reviewers’ concerns as follows.
Regarding the limitation of single dataset (R1, R3, & R4): We acknowledge that multicenter validation would improve the feasibility and mitigate biases in our approach for clinical practice. However, gathering PET images from Parkinson’s disease (PD) patients is difficult due to instrument constraints, particularly for CFT PET scans. Indeed, we are actively collaborating with additional hospitals to gather pertinent data for validating the effectiveness of our MetaAD.
Regarding the reproducibility of the model (R1, R3, & R4): We recognize the significance of data transparency and reproducibility in scientific research. We apologize for the omission of the link to relevant code, which is restricted by the double-blind review process. We commit to making our code publicly accessible once our paper is finally accepted. As for the dataset, however, we cannot make it publicly available due to the confidentiality agreements and privacy concerns.
Regarding the improvements of paper writing (R1 & R3): Thank you for your thorough review. We apologize for any occasional spelling and citation errors in the manuscript. We will address these issues and strive for improvements in the final version of the paper, including correcting typos, polishing texts, and enhancing the organization.
Regarding the comparison with more modern pipelines (R3): Thank you for your feedback. In the current era of AIGC, leveraging cutting-edge frameworks like diffusion models for image generation seems like a direct approach. However, this does not apply to our task due to the significant challenge of synthesizing unpaired 3D images. Unpaired image translation remains a challenge for diffusion models. While some recent works have attempted to use diffusion models for unpaired image translation (e.g., CycleGAN-Turbo, Gaurav Parmar, et al, 2024), they are typically based on Stable Diffusion pre-trained on natural images and primarily focus on 2D images, which is not suitable for unpaired 3D medical images.
Regarding the details of the limitations and future work (R3): While our current work is intriguing, it is still in a preliminary stage, thereby leaving considerable scope for further exploration. Due to the limited page length, we only briefly highlight future research directions in our paper, which aligns with ongoing efforts and have yielded promising results. In the final version, we intend to elaborate more comprehensively on these points if additional page is available.
Regarding the analysis of bad performance of other UAD methods (R4): Thank you for your interest. As we elaborated in the latter part of the introduction, many UAD methods aim to learn a compact latent space exclusively representing the normal data distribution. However, optimizing this latent space poses a significant challenge. Conventional reconstruction networks may encounter an “identical shortcut” where both normal and anomalous samples are well-recovered, leading to a failure in detecting outliers effectively. To overcome this challenge, we substitute the latent space with CFT images, facilitating easier and more efficient optimization. Additionally, we introduce Abnormal Metabolism Suppression to further mitigate “identical shortcut” in the cyclic modality translation workflow, while conventional CycleGAN without it might still suffer.
Meta-Review
Meta-review not available, early accepted paper.