List of Papers Browse by Subject Areas Author List
Abstract
Quantitative analysis of brain iron is widely utilized in neurodegenerative diseases, typically accomplished through the utilization of quantitative susceptibility mapping (QSM) and medical image registration. However, this approach heavily relies on registration accuracy, and image registration can alter QSM values, leading to distorted quantitative analysis results. This paper proposes a multi-modal multitask QSM reconstruction algorithm (mQSM) and introduces a mutual Transformer mechanism (mTrans) to efficiently fuse multi-modal information for QSM reconstruction and brain region segmentation tasks. mTrans leverages Transformer computations on Query and Value feature matrices for mutual attention calculation, eliminating the need for additional computational modules and ensuring high efficiency in multi-modal data fusion. Experimental results demonstrate an average dice coefficient of 0.92 for segmentation, and QSM reconstruction achieves an SSIM evaluation of 0.9854 compared to the gold standard. Moreover, segmentation-based (mQSM) brain iron quantitative analysis shows no significant difference from the ground truth, whereas the registration-based approach exhibits notable differences in brain cortical regions compared to the ground truth. Our code is available at https://github.com/TyrionJ/mQSM.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2723_paper.pdf
SharedIt Link: pending
SpringerLink (DOI): pending
Supplementary Material: https://papers.miccai.org/miccai-2024/supp/2723_supp.pdf
Link to the Code Repository
https://github.com/TyrionJ/mQSM
Link to the Dataset(s)
https://github.com/TyrionJ/mQSM
BibTex
@InProceedings{He_mQSM_MICCAI2024,
author = { He, Junjie and Fu, Bangkang and Xiong, Zhenliang and Peng, Yunsong and Wang, Rongpin},
title = { { mQSM: Multitask Learning-based Quantitative Susceptibility Mapping for Iron Analysis in Brain } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15002},
month = {October},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors combined ResNet and U-Net to jointly perform brain QSM and segmentation. T1 is required, but in clinical practice T1 is almost always acquired in any brain protocol.
- 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.
Their joint reconstruction (QSM and segmentation) provide a way to get away with registraion. If “whole-brain iron analysis” is performed, the proposed method would be of great interest to clinical/research communities.
- 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.
-
Major concern is related to the motivation of the work and its relevance to clinical practice. The whole brain iron analysis based on QSM is not theoretically feasible not because of registration but susceptibility anisotropy that scalar QSM cannot address - STI may be needed. ROI-based analysis focused on deep gray for instance is what is clinically more relevant.
-
There are many unsupported claims. For example, “lightweight and efficient multi-modeal data fusion mechanism”, “precise all-in-one solution”, and “significantly boosting their performance”. These claims need to be supported by either theoretical or experimental results.
-
QSM producer is introduced, but its justification and analysis are completely missing as this seems to play a crucial preprocessor of the ResNet.
-
Lack of innovation/novelty: after all this is a parellel combination of a resnet and U-net.. What’s new here that is of interest to the MICCAI readership?
-
Evaluation of segmentation accuracy of the proposed method is missing. Therefore, the subsequent analysis in Section 3.3 is concerning. Change in QSM through registration to atlas is obvious.. The author’s claim on this is not surprising, but as far as I am concerned, this is not what is clincial performed, again QSM has not and would not probably be used for “whole-brain iron analysis” due to its theoretical limitation.
-
- Please rate the clarity and organization of this paper
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?
The provided link (https//anonymous.site.link) in the Abstract seems broken.
- 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
Please see weaknesses.
- 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?
If the authors can clarify the current clinical practice in terms of QSM, whole-brain iron analysis based on QSM, and registration, and if the authors’ claims are supported, I may consider changing my recommendation.
- 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
When conducting analysis from QSM, it typically starts with ROI selection. However, manual ROI selection has its limitations, and segmenting solely based on QSM is challenging. Therefore, the authors proposed a method to integrate multi-modal information by appropriately using T1 and QSM, while also performing QSM reconstruction. Through this approach, the authors claimed to provide a precise all-in-one solution for QSM.
- 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 mTrans framework addresses the limitations associated with inappropriate ROIs.
- This paper proposes a method for effectively and efficiently utilizing T1 and QSM by appropriately fusing the two pieces of information.
- The study conducted comparisons with existing QSM 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.
-
I felt that there is a lack of sufficient or clear explanation regarding the problem in both the introduction and abstract. It seems that supplementation is needed.
-
There is insufficient description of the data. QSM varies depending on the disease group. This can have a significant impact on the results, and clarification is necessary.
-
- Please rate the clarity and organization of this paper
Satisfactory
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- Do you have any additional comments regarding the paper’s reproducibility?
I didn’t see executable links included in the submission. Additionally, there isn’t sufficient explanation about the subjects, making it unclear to which extent the method can be applied to different groups.
- 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
- Have you conducted comparisons with commonly used tools such as FSL and Freesurfer for obtaining masks through T1?
- Have you attempted training after reconstructing QSM using different methods from the QSM producer?
- Were the subjects composed of healthy individuals? Have you also verified the results from individuals with diseases?
[MINOR]
- Figure 2 and Figure 3 is difficult to interpret with captions alone. Therefore, providing additional explanations would be helpful. For example, including legends with explanations for each arrow would enhance understanding.
- In Fig3, the differences in result are subtle. Adding zoom images to highlight the differing areas would be beneficial for clarity.
- There is a typo in 2.1. “kernell” should be corrected to “kernel.”
- 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?
Creating ROI mask is a important step in analyzing images like QSM. Author approached this process efficiently by fusion methods across T1 and QSM.
- 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
Authors propose a precise all-in-one solution for QSM reconstruction and brain iron analysis with a lightweight and efficient multi-modal data fusion mechanism named mutual Transformer (mTrans), which is also able to plug in to other models to signicantly boost their performance. Their approach avoids registration induced QSM value changes that accumulate errors. Experiemental results are competitive to baselines.
- 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.
-
Image registration as a pre-processing step accumulates errors that induce to bias in the following analysis. The idea of avoiding registration is a good angle and leading to efficient approach.
-
QSM reconstruction results by mQSM have the highest PSNR and NRMSE. In the mean time, results by mQSM retain the raw brain iron quantatives.
-
Ablation studies show the effectiveness of the proposed modules.
-
- 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 table 1, STAR-QSM has the highest SSIM instead of mQSM, which is incorrectly bolded.
-
Table 3 and Fig 4 have a different font format with other tables in the paper.
-
Authors claim the proposed module, mTrans, is able to be applied to other models. However, I didn’t see the experimental results of other models using mTrans.
-
Code link in abstract is unavailable.
-
- 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
Good paper in total.
- 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?
Given the good idea and sufficient experiments in this paper, I’m happy to see it accepted.
- 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
Author Feedback
To Reviewer #1 1.Thank you for mentioning of STI. According to current research, there are deep learning-based methods (10.1016/j.neuroimage.2021.118376, 10.3389/fnins.2022.837721) using single scan to preserve susceptibility anisotropy. STI requires echo data from six different directions for calculation, making it impractical for clinical applications. Therefore, this study uses QSM as a non-invasive method for assessing brain iron. A significant clinical application is the use of segmentation methods instead of registration to improve the accuracy, as demonstrated by the results in Fig. 3, Tab. 2, and the supplementary materials. 2.Thank you for raising concerns about some claims. A highlight of this study is the multi-task QSM reconstruction, including reconstruction and segmentation. Having these two results allows for brain iron quantification immediately, constituting an “all-in-one solution.” The precision lies in the above mentioned that segmentation-based methods are more accurate than registration-based. The results from the ablation experiments show that the proposed multimodal fusion method indeed enhances model performance. 3.Thank you for pointing out the completeness issues. The QSM reconstruction primarily relies on the TKD method. The justification and analysis of this approach are supported by research on TKD. Therefore, a detailed discussion of the TKD method is not included in this study. 4.Thank you for your concerns about the innovations. The innovations are as follows: it’s the first to propose the concept of multi-task QSM reconstruction, offering readers new insights in QSM reconstruction. This study demonstrates that segmentation-based methods are more effective than registration-based methods for brain iron quantification. This finding, in turn, provides a novel perspective for research in the neurological field. 5.Thank you for your insights on the evaluation of segmentation. This study validates that the QSM quantification of brain regions based on segmentation results shows no significant difference from the ground truth, indirectly reflecting the accuracy of the segmentation results. Regarding the clinical application of QSM, numerous studies in Radiology have reported the use of QSM for brain iron quantification and its application in neurodegenerative diseases. QSM has already been incorporated as a routine examination sequence, providing a non-invasive method for assessing brain iron distribution. Additionally, recent research has introduced sub-voxel QSM methods, allowing for a more detailed analysis of the distribution of paramagnetic and diamagnetic substances. This advancement offers tools for both brain iron quantification and demyelination analysis.
To Reviewer #3 1 & 2. Thank you for pointing out these issues, which will be addressed in the revised version.
- The ablation experiments validated that mTrans can be directly applied to U-Net models with self-attention mechanisms. However, for other types of models, further experiments are indeed necessary to verify its applicability. This will be addressed in future work.
- Due to anonymity requirements, a fake link was used.
To Reviewer #4
- I apologize if the explanation in the paper did not meet your expectations. The focus of this study is on investigating multi-task QSM reconstruction algorithms. It proposes that for quantitative analysis using QSM, segmentation-based methods are more accurate than registration-based methods.
- Thank you for providing feedback on the data description. The QSM data utilized are sourced from normal volunteers, patients with AD, and patients with cerebral small vessel disease (in the revised version). Due to the versatility of QSM in clinical applications, the research methodology of this paper is not limited to a specific disease. [MINOR] Thank you for your feedback on the figure captions and some spelling errors. These will be corrected in the revised version.
Meta-Review
Meta-review not available, early accepted paper.