List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Ying-Qiu Zheng, Harith Akram, Stephen Smith, Saad Jbabdi
Abstract
The ventral intermediate nucleus of thalamus (Vim) is a well-established surgical target in magnetic resonance-guided (MR-guided) surgery for the treatment of tremor. As the structure is not identifiable from conventional MR sequences, targeting the Vim has predominantly relied on standardised Vim atlases and thus fails to model individual anatomical variability. To overcome this limitation, recent studies define the Vim using its white matter connectivity with both primary cortex and dentate nucleus, estimated via tractography. Although successful in accounting for individual variability, these connectivity-based methods are sensitive to variations in image acquisition and processing, and require high-quality diffusion imaging techniques which are often not available in clinical contexts. Here we propose a novel transfer learning approach to accurately target the Vim particularly on clinical-quality data. The approach transfers anatomical information from publicly-available high-quality datasets to a wide range of white matter connectivity features in low-quality data, to augment inference on the Vim. We demonstrate that the approach can robustly and reliably identify the Vim despite compromised data quality, and is generalisable to different datasets, outperforming previous surgical targeting methods.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43996-4_17
SharedIt: https://rdcu.be/dnwOR
Link to the code repository
https://git.fmrib.ox.ac.uk/yqzheng1/hqaugmentation.jl
https://git.fmrib.ox.ac.uk/yqzheng1/python-localise
Link to the dataset(s)
https://www.humanconnectome.org/study/hcp-young-adult/data-releases
https://www.ukbiobank.ac.uk/enable-your-research/about-our-data/imaging-data
Reviews
Review #1
- Please describe the contribution of the paper
The authors presented a transfer learning based approach to target VIM, in particular leveraging high quality datasets in order to obtain better anatomical information on lower quality clinical data. The study performed analysis on accuracy, generalization and reliability of the approach and reported better performance than previous surgical targeting methods.
- 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.
- Accurate surgical targeting, e.g. of the VIM, is critical in delivering desired and optimal treatment for patients suffering from tremor and/or other forms of movement disorder. This work is an interesting contribution to this overall effort.
- The concept of using high quality data for transfer learning purpose to better assist anatomical information delineation and extraction from low quality clinical data is interesting and neat.
- The paper is well organized and presented. The figures are well done. And the authors should be commended for their effort in putting together additional content of the work in supplemental material format.
- It appears the authors would make this code available; that initiative should be commended.
- 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 major concern is data presentation in this paper. While the figures in the paper illustrate some understandings of the performance with respect to the metrics outlined, i.e. Dice and centroid displacement, the paper lacks any numerical representation of these metrics to allow proper assessments of the performance of the approach, especially its clinical usefulness. It’s quite impossible to assess average, standard deviation, as well as best and worst scenarios from the figures. Numerical analysis and presentation of the data will further support the paper’s conclusions in addition to visual impression via figures currently, as well as potential statistical test to indicate significance of the conclusions.
- 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 work should be reproducible with reasonable efforts, especially as the authors indicated that multiple key items including the code would be made public.
- 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
- It would be interesting to examine other structures such as STN as VIM is not the only target of interest for e.g. DBS therapies for movement disorders.
- It would be interesting to also examine Jaccard index as another tool or metric to Dice.
- 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 recommendation is based on the lack of numerical data presentation that hinders proper assessment of the performance of the approach.
- 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
4
- [Post rebuttal] Please justify your decision
The authors’ feedback and responses to comments are appreciated, and they address some of the concerns. While acknowledging and adding additional quantitative results in supplemental materials are recognized, it is difficult to assess the quality of the results of the work, subsequently the impact of the work to the field, in its current form.
Review #2
- Please describe the contribution of the paper
The work tackles a clinically important question: to provide patient-specific target maps for the Vim nucleus of the thalamus for image-guided neurosurgical procedures. To do so, the authors utilize DT-MRI and connectivity information in low-resolution (clinical) data. Their method transfers knowledge from publicly available, high-resolution DTI datasets to improve the accuracy of connectivity estimates in the low-resolution data. Their method can robustly and reliably identify the Vim, and was proven to be generalizable to different datasets, outperforming previous, related targeting methods.
- 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.
- Important clinical problem
- DTI may carry subject-specific information on thalamus anatomy, allowing better targeting
- Overcomes some limitations of the angular resolution of clinical DTI data
- Approach relying on public datasets
- 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 registration accuracy of dMRI data to the structural images is critical and has not been addressed
- The authors basically validate their method against Atlas based approaches, which is quite limited in terms of adapting the anatomy to the subject’s anatomy
- Lack of ground truth
- Little discussion/interpretation of what the improvement accuracy means in 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 authors relied on some public datasets and transfer knowledge to a different domain, so as far as I can judge, it might be reproducible for other group’s data.
- 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
One limitation is that the registration accuracy between dMRI and structural images is a crucial factor that has not been adequately addressed. The authors validate their proposed method by comparing it to Atlas-based approaches, which have limitations in terms of adapting the anatomy to the subject’s unique anatomy. This suggests that the proposed method may be more effective in accounting for individual differences in anatomy. Additionally, there is little discussion or interpretation of what the improvement in accuracy means in practice, which limits the practical implications of the study’s findings. The main hypothesis is that connectivity information allows the localization of the Vim, or at least a cluster of connectivity that outlines mainly the specific motor nuclei of the thalamus. This method relies clearly on subject-specific imaging information, however, the inherent spatial resolution limitation of dMRI is prevalent even if the angular resolution and specificity is improved through the transfer learning method from high-resolution datasets. While the method is very interesting and innovative, it is not clear what the ground truth really is – in my opinion, the clinical transferability would certainly necessitate more thorough validation, such as by performing high-resolution MRI with thalamus specific contrast, histology, electrophysiology (e.g. DBS placement). in the same dataset.
- 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?
It is a topic that’s rather unique for the MICCAI - could be a great contribution to the field of image guided neurosurgery. Despite some weaknesses, it is a well written paper. I believe the weaknesses could be addressed in a minor revision.
- Reviewer confidence
Somewhat confident
- [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 #4
- Please describe the contribution of the paper
This paper proposes a transfer learning methods named HQ-augmentation model to detect and segmentation the ventral intermediate nucleus of thalamus (Vim) from low-quality diffusion MRI. The method focuses on the migration of Vim features learned at high-quality images to the low-quality at the feature level.
- 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 transfers Vim features from high-quality images to low-quality medical images at feature level
- The method is applied in Vim’s detection and segmentation in low-quality and achieved good results, suggesting its potential for translation into a reliable clinical tool.
- 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) The created surrogate low-quality (LQ) datasets from the high-quality dataset should be described more detailed. And How to guarantee the similarity between the generated low-quality dataset and the real data in the clinic ? (2) More experiments should be conducted to verify the robustness of the method. For instance, some studies pay attention to synthesising high-quality images from low-quality, such as 7T from 3T. Is it possible to use synthesis methods to generate high-quality MRI and segment the Vim? (3) The figures in the manuscript are not clear enough and difficult to read. For example, the arrows and squares in Figure 1 should be described in more detail
- 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 model is difficult to be reproduced.The model setup and training details are not decribed, which can make the work less reproducible.
- 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 in the 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
5
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The method is novel and results are promising, however there are several concerns as mentioned in the weakness section.
- Reviewer confidence
Somewhat confident
- [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.
This paper addresses a relevant clinical problem of targeting the Vim on clinical-quality data and proposes a novel perspective of transferring anatomical information from high-quality data. There are also a few weaknesses that need to addressed. For example, the statistics need to be summarized, the limitation of the methodology and evaluation should be discussed, and the clinical implication of the improved accuracy should be elaborated.
Author Feedback
Thanks for the feedback. The reviewers all agreed that the idea of leveraging high-quality images to facilitate surgical targeting on low-quality images is novel and clinically important, that the paper is well organised, and that the results are encouraging. As highlighted by R2, our paper is rather unique within MICCAI and could be a great contribution to the field of image-guided surgery. Below we address the main concerns. 1.R1 on the lack of statistics: In our initial presentation of results, we opted for scatter and violin plots, aiming to provide a visually informative summary of data distribution, a common practice in neuroimaging studies. Acknowledging your suggestion, we will include additional statistical tests and the suggested Jaccard index in the supplement for a more comprehensive data presentation. 2.R2 on insufficient validation (limitation of evaluation): we’d like to clarify that our validation process was twofold. We validated our results not only against an atlas but also against the individual-specific Vim, derived from high-quality dMRI using its primary connectivity properties. The latter, referred to as the connectivity-driven Vim, has been previously validated in the literature as predictive of surgical outcomes, and thus, can serve as a form of ground truth. This dual-layered validation lends robustness to our methodology and findings. 3.R2 on lack of ground truth/inherent limitation of dMRI (limitation of methodology): we agree that this is an important aspect of our study and relates to the previous point. The absence of an absolute ground truth in this field is indeed a prevalent challenge. Nevertheless, our approach uses individual-specific Vim, derived from high-quality dMRI, as a proxy for ground truth. The high-resolution dMRI data we used for deriving this “ground truth” have an isotropic spatial resolution of 1.25mm, which is sufficiently detailed for our purposes, given that the Vim cluster typically measures around 4x4x6mm. Thus, our approach, while not flawless, gains significant credibility by validating against this proxy for ground truth, helping to counter the limitations of our methodology. 4.R2 on clinical implication of improved accuracy: In current DBS procedures, one or two electrodes are placed near established targets. These electrodes have multiple contact points to increase the likelihood of achieving beneficial outcomes without severe side effects. Enhanced accuracy implies a greater overlap with the target area, thereby increasing the chances of successful surgical outcomes. Our methodology is not restricted to the Vim; it can be applied to any DBS target. We plan to release a preoperative tool that maximises successful targeting on an individual basis, while minimising patient burden. Hence, the potential clinical impact is substantial. 5.R2 on registration accuracy between dMRI and T1: We promise to address this aspect in the supplement. 6.R4 on whether surrogate low-quality data was representative of real data. The process of creating these surrogate low-quality datasets involved careful consideration to ensure their clinical relevance. Varying number of shells and spatial resolution were adopted to emulate common clinical data protocols. We promise to provide more details in the supplement. 7.R4 on segmenting Vim from synthesised high-quality images: we are not aware of techniques that can generate high-quality “connectivity” maps from low-quality data. This is crucial here, as the Vim is defined through its connections, because even high-quality T1 or DTI images lack the contrast to find Vim. In addition, even if there was a method that could generate high-quality diffusion data using synthesisation, this would still not be as robust as our method, as we have found that the Vim segmentation fails frequently even in high-quality diffusion data. Our approach of quality transfer of the connectivity maps, as opposed to the raw data, directly addresses this problem.
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 authors plan to add the statistics, which I believe is achievable. The limitation of validation is properly discussed. The clinical implication is well explained. Overall, I think the paper addresses the major issues and can be accepted.
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.
This paper proposed a patient-specific framework to detect and segmentation the ventral intermediate nucleus of thalamus. The clinical significance is strong and the overall pipeline is easy to follow. Although one reviewer argued the technical innovation of this work, I still recommend the acceptance of this work due to considering the clinical impact.
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 is quite an interesting and original approach to treating tremor near ventral intermediate nucleus of thalamus, as this is an often unaddressed problem in image-guided surgery. I believe the paper could be of interest for the majority of people in the field, and the paper is generally well written. While I agree with R1 this work could mature further with additional analyses, for MICCAI I tend to believe this should be good enough for acceptance.