List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Henry F. J. Tregidgo, Sonja Soskic, Mark D. Olchanyi, Juri Althonayan, Benjamin Billot, Chiara Maffei, Polina Golland, Anastasia Yendiki, Daniel C. Alexander, Martina Bocchetta, Jonathan D. Rohrer, Juan Eugenio Iglesias
Abstract
The human thalamus is a subcortical brain structure that comprises dozens of nuclei with different function and connectivity, which are affected differently by disease. For this reason, there is growing interest in studying the thalamic nuclei in vivo with MRI. Tools are available to segment the thalamus from 1 mm T1 scans, but the image contrast is too faint to produce reliable segmentations. Some tools have attempted to refine these boundaries using diffusion MRI information, but do not generalise well across diffusion MRI acquisitions. Here we present the first CNN that can segment thalamic nuclei from T1 and diffusion data of any resolution without retraining or fine tuning. Our method builds on our histological atlas of the thalamic nuclei and silver standard segmentations on high-quality diffusion data obtained with our recent Bayesian adaptive segmentation tool. We combine these with an approximate degradation model for fast domain randomisation during training. Our CNN produces a segmentation at 0.7 mm isotropic resolution, irrespective of the resolution of the input. Moreover, it uses a parsimonious model of the diffusion signal (fractional anisotropy and principal eigenvector) that is compatible with virtually any set of directions and b-values, including huge amounts of legacy data. We show results of our proposed method on three heterogeneous datasets acquired on dozens of different scanners. The method is publicly available at freesurfer.net/fswiki/ThalamicNucleiDTI.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43993-3_24
SharedIt: https://rdcu.be/dnwNp
Link to the code repository
https://github.com/htregidgo/joint_diffusion_structural_seg
Link to the dataset(s)
https://www.humanconnectome.org/
Reviews
Review #2
- Please describe the contribution of the paper
The authors present a new segmentation method for the components of the thalamus. The basic idea is to perform a lot of augmentations to make the approach robust to resolution changes, which seems to work.
- 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.
Enables good segmentations even on low resolution images where state of the art methods fail.
- 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.
Minor weaknesses:
- no significance testing was done
- pease use boxplots, the table is no nice to look at
- 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 state that the tool will be made openly available.Data is also available, so the paper should be 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
-
- 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?
This is a method that makes a good thalamic nuclei segmentation possible on a much broader range of datasets and could therefore have a real impact in the field.
- 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 author claims to present the first CNN that can segment thalamic nuclei. Their model is image resolution invariant. They extensively discuss the steps involved in their method. they experiment with HCP, LOCAL, and ADNI 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.
The main strength of the paper is to provide resolution-invariant method to segment thalamic nuclei. They test and train on different 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 authors claim they present the first CNN which is a little exaggeration since they did not mention “Convolutional Neural Network Based Frameworks for Fast Automatic Segmentation of Thalamic Nuclei from Native and Synthesized Contrast Structural MRI” paper which is also CNN-based and solves the same problem. Umapathy, Lavanya, et al. “Convolutional Neural Network Based Frameworks for Fast Automatic Segmentation of Thalamic Nuclei from Native and Synthesized Contrast Structural MRI.” Neuroinformatics (2021): 1-14. In terms of method, the novelty was not clear because standard UNet was used.
- 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
They provide the ready-to-use tool, so reproducible and helpful.
- 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 author should include a discussion on previous studies, “Umapathy, Lavanya, et al. “Convolutional Neural Network Based Frameworks for Fast Automatic Segmentation of Thalamic Nuclei from Native and Synthesized Contrast Structural MRI.” Neuroinformatics (2021): 1-14.”.
- the author should include a discussion on novelty of the method not in terms of application which would be to apply CNN.
- In Table 1: it was not clear why would the score be greater for the model trained on manual annotations while for manual annotations the score is low.
- It seems the method performs similar to that of FSL, so I was wondering why do we need this method. Author should demonstrate what we cannot do using FSL and what is new here.
- The time taken to segment should be mentioned, if it could be used in near-real time tasks.
- 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 paper was written clearly and the problem that they are trying to solve is interesting. The novelty of the method and application is in question. The justification regarding the method contribution and the paper that I mentioned before would be helpful.
- 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
The paper proposes novel CNN based segmentation method for T1/DTI data with different resolutions. The main contributions are in the generation of training data with domain randomization and various data augmentation strategies. Experimental results on three datasets with various resolutions were presented to show improved performance as compared to existing Bayesian based segmentation in FreeSurfer.
- 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.
Domain-agnostic segmentation of thalamic sub-nuclei are valuable and holds the promise of applicable DTI data in clinical studies.
- 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 reduction of DTI data to a tensor representation makes the proposed method generally applicable to legacy data, it is not optimal for modern diffusion MRI data including those from ADNI.
Technical innovation in terms of deep learning is limited given that the main effort is on data augmentation.
Limited evaluation on DTI data of clinical quality.
- 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 stated the codes will be distributed publicly.
- 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 DTI and T1 data are not guaranteed to align perfectly given the distortion artifacts in DTI. The segmentation accuracy can be affected by the variable distortion across subjects.
The clinical value of the subnuclei segmentation method is not sufficiently demonstrated.
- 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?
This work proposes a thalamic subnuclei segmentation network that can potentially be applied to DTI data of arbitrary resolution. This can be quite valuable for various clinical studies.
- Reviewer confidence
Very 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 work addresses a problem that has been of interest to parts of the MICCAI community for many years, and reviewers thought that it presents an approach that, even though its technical novelty is somewhat limited and appears to focus on a well-engineered data augmentation strategy, produces results that are strong enough that it should provide a practical benefit. Even though I do not think a rebuttal is required, authors are encouraged to account for the constructive feedback from the reviewers as much as possible in their final version.
Author Feedback
We would like to thank the reviewers and meta-reviewer for their kind assessment of our paper. We apologise that the limited space available for this submission restricted our ability to display results graphically, rather than in tables, and to demonstrate the utility of our tool in specific research problems outside of the limited Alzheimer’s discrimination study. We are already planning further work using this tool to investigate the nuclei affected by different forms of genetic frontotemporal dementia.
Following the comments from the reviewers, we have included a short discussion item (within space constraints) on the impact of potential misregistration between the DTI and T1 data, due to geometric distortion in the former – which, in general, is more problematic in frontal and occipital regions. We have also added statistical testing on our Alzheimer’s discrimination experiment, including a statement of which nuclei reach significance. Finally, we have also clarified that our method is the first thalamic segmentation network for combined diffusion and structural MRI and that the lack of a retraining requirement for new acquisitions is novel compared to the existing T1 only network (Umapathy et al 2021), which has been shown to fail on images from specific MRI manufacturers.
We would also like to clarify that no manual annotations were used for training, and we hypothesise that improvements in Dice scores for our network over the Bayesian implementation is due to the combination of three Bayesian segmentation models to generate our training set. Additionally, the major advantage of our method is the delineation of individual nuclei comprising the thalamus, which are affected differently by disease and involved in separate functions. This is compared to standard segmentation pipelines such as FSL’s “FIRST” and FreeSurfer’s recon-all which only segment the exterior boundary of the thalamus. Finally, while inference time for our method is only a few seconds per subject when run on the GPU, acquisition time and pre-processing steps required for DTI make near-real time applications currently infeasible.