List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Athena Taymourtash, Hamza Kebiri, Ernst Schwartz, Karl-Heinz Nenning, Sébastien Tourbier, Gregor Kasprian, Daniela Prayer, Meritxell Bach Cuadra, Georg Langs
Abstract
Resting-state functional Magnetic Resonance Imaging (fMRI) is a powerful imaging technique for studying functional development of the brain in utero. However, unpredictable and excessive movement of fetuses have limited its clinical applicability. Previous studies have focused primarily on the accurate estimation of the motion parameters employing a single step 3D interpolation at each individual time frame to recover a motion-free 4D fMRI image. Using only information from a 3D spatial neighborhood neglects the temporal structure of fMRI and useful information from neighboring timepoints. Here, we propose a novel technique based on four dimensional iterative reconstruction of the motion scattered fMRI slices. Quantitative evaluation of the proposed method on a cohort of real clinical fetal fMRI data indicates improvement of reconstruction quality compared to the conventional 3D interpolation approaches.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16446-0_57
SharedIt: https://rdcu.be/cVRT0
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #3
- Please describe the contribution of the paper
The authors attempt to address the motion1. challenges in fetal MRI by looking at the spatio-temporal dimensions of the 4D fMRI data set as a whole rather than as a set of 3D volumes over time.
- 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 spatio-temporal formulation is intuitive and makes sense
- The use of well-established regularizer such as TV makes it easy to implement and reproduce
- 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 computational effort of this algorithm needs to be detailed - specifically memory and time per iteration
- The choice of the Lagrange parameter (Alphas in the current implementation) needs to be demonstrated based on L-curves or other methods - Tikhonov for example.
- It will be useful to see the original SSIM values instead of just the difference
- 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 authors meet the reproducibility criteria. The reviewer was unable to find any code or a commitment to share later.
- 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/2022/en/REVIEWER-GUIDELINES.html
The authors have pursued an important problem - motion mitigation in fetal fMRI. The strengths and weaknesses are outlined above.
- 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?
The problem of fetal fMRI motion is well motivated. Looking at the spatio-temporal data as a whole is an intuitive approach. The use of TV is simple to implement and reproduce. The details related to computation - time, effort and convergence - needs to be discussed.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
2
- 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 #4
- Please describe the contribution of the paper
As a general preprocessing step of fetal fMRI analysis, this manuscript proposes a four dimensional (4D) iterative reconstruction method to correct the motion scattered fMRI slices, while existing methods processing 3D images at each time frame individually.
- 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.
Motivation of the proposed is well illustrated. The presented method is well presented and easy to follow. Experiments were performed on real clinical fetal fMRI data. Down-steam tasks were used to measure the effect of motion compensation (Section 3.4).
- 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.
Comparison is not fair enough from my point of view. In Section 3.3, baseline methods are simple interpolation of independently re-aligned 3D images of different time frames. Firstly, comparing the proposed method with linear, cubic, and sinc interpolation is redundant. There is not a significant gap between different interpolation methods, thus it is not very necessary to compare all this interpolation method. Also, better baseline methods, rather than simple interpolation, would make the proposed method more convincing. Secondly, for fMRI analysis, is it necessary and important to interpolate between different time frames? It seems interpolation is not used in experiment in Section 3.4 (Fig. 4). In Section 3.4, it seems that the proposed method is compare with observed image without any re-alignment, which is not fair enough. It is better to perform basic 3D registration before fMRI analysis. If I misunderstand something in Section 3.4, this should be clarified.
- 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 proposed method seems to be reproduceable. The method is clearly presented and there is an algorithm summarizing the workflow.
- 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/2022/en/REVIEWER-GUIDELINES.html
My major concern is the comparison of baseline methods. Authors only compare their proposed method with some simple interpolation-based method, which can be much improved in my point view. Fair comparison with baselines would make this manuscript more convincing. Check the ``main weakness’’ for more details.
While This manuscript is well organized and easy to follow, there are still several things unclear to me. What does the metric ‘‘L’’ mean in Fig. 3 (a)? I do not find any definition all through the manuscript. Also, SSIM metrics require a reference image, but obvious there is no such a ‘‘ground-truth’’ in this dataset. Fig. 2. shows a Reference volume, but the main text does not have any description of the reference volume. Also, the ‘‘s2v’’ in Fig.2 is not properly abbreviated.
Language-related errors that can be easily fixed. ‘‘…, whereas the proposed 4D iterative reconstruction did recover the entire brain.’’
- 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 motivation, method part are very well, but the experiment part can be much improved.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
3
- 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 #5
- Please describe the contribution of the paper
The authors propose a spatial-temporal iterative reconstruction method for motion correction of in-utero fetal fMRI. Experimental results on a cohort of real clinical fetal fMRI data indicate improvement of reconstruction quality compared to the conventional interpolation approaches.
- 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.
Novel reconstruction of a 4D image from a single sequence acquired over time by employing the spatial-temporal similarity Solid evaluation in real clinical data
- 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 iterative spatial-temporal reconstruction may be computationally expensive; the computation time should be clarified The motion is estimated prior to the iterative reconstruction. It is unclear whether the motion parameters can also be iteratively updated during the reconstruction.
- 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 algorithm is clearly explained, though the data and code are not made available.
- 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/2022/en/REVIEWER-GUIDELINES.html
1: Iterative reconstruction of the fMRI data with 96 volumes using the low-rank and total variation constraints can be computationally demanding. Please comment on the computation complexity. 2: How the downsampling and blurring operators in the reconstruction equation are designed should be clarified.
- The motion operator is missing in Eq. [7].
- In the current algorithm, the motion is estimated prior to the iterative reconstruction. The curiosity is whether the motion parameters can also be iteratively updated during the reconstruction.
- 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?
Novel motion-corrected reconstruction of 4D in-utero fetal fMRI by employing the spatial-temporal similarity, which results in much better image quality than non-iterative interpolation methods.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
2
- 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 authors propose a framework for addressing motion artifacts in fetal brain MRI images via spatio-temporal motion correction as well as image reconstruction. This is an important problem is medical image analysis and of great interest to the MICCAI community.
The submission is clearly written, well motivated and experiments are run on real clinical data sets.
A summary of comments and requests from reviewers are as follows:
The authors should include citations for previous methods that have applied similar methods to address the proposed problem.
Although a well established regularizer is proposed, an algorithm summary is presented and the method is clearly described, no (reference to) code and/or data was found regarding reproducibility.
Reviewers have identified missing methodological details in the submission, such as computational burden, memory requirements, full SSIM values. It has also been raised that in the experimental section, the baseline methods (3D interpolation approaches) should have been chosen to be more challenging ones to make the comparison with the proposed one more fair. Additionally, clarifications in Sec 3.4, Functional connectivity analysis, would be welcome.
- What is the ranking of this paper in your stack? Use a number between 1 (best paper in your stack) and n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
1
Author Feedback
We thank all reviewers for their comments and constructive feedback. We will submit a revised version of the paper and update the main text for the typos and language-related mistakes. Here we try to address your concerns and questions as much as possible: Computational cost: All programs were run in Linux environment on a standard PC using a single thread of an Intel(R) Xeon(R) CPU (E5-2620 @ 2.10GHz). Interpolation methods are computationally efficient as it takes about 5 mins for Linear and about 12 mins for Spline and Sinc for one loop over a full 4D image. The average convergence time of the proposed iterative method is about 3 hours and 40 mins per subject. Optimization parameters: The choice of regularization parameters was determined empirically by visual assessment of the L-curves but the choice of the alpha parameters was based on the notion of equally important dimensions. Unfortunately, these results cannot be included in the paper due to space limitation. SSIM index: It was measured between all possible pairs of 3D volumes in fMRI time series (96*95/2=4560) and the average value was reported for each subject. Original values of SSIM, SD, and L (Laplacian, sharpness) will be added to Table.1 of the Supplementary Information as suggested. Baseline methods: To the best of our knowledge, 3D linear interpolation is widely used for the reconstruction of motion-corrected in-utero fMRI time series and state of the art algorithms mainly focused on improving the estimate of realignment parameters (for potentially large movement of fetuses) than reconstruction. A recently published paper (July 2022) has suggested iterative reconstruction for each 3D volume separately based on the super-resolution technique, we will add citation to it in the final version of the paper and the comparison will be the subject of our future works. Interpolation between fMRI time frames: We didn’t interpolate between different time frames and 3D interpolation approaches were applied to each 3D volume separately. Our proposed method doesn’t interpolate between volumes as well but given that neural activity is characterized by low-frequency oscillations of the BOLD signal, we integrated the desired feature of temporal smoothing to the reconstruction algorithm by four-dimensional modeling and regularization. Figure 2: Reference volume for each subject was generated by finding automatically a set of consecutive volumes of fetal quiescence and averaging over them as suggested in reference 13. In summary, we first performed gross motion estimation for each volume and then we selected an arbitrary 3D point and applied each of the frame-wise transformations to this point to get a point trajectory. Distance between consecutive points in the trajectory were calculated and a moving average of a preselected number (in our experiments we use 10) was computed. The set of frames corresponding to the minimum value of the moving average was selected and averaged. Now that a high-quality reference volume has been computed, all slices from all frames were iteratively registered to it (slice-to-volume registration, s2v) by maximization of normalized cross-correlation with respect to 6 rigid parameters. An example of the achieved parameters and the reference volume was shown in figure 2. Downsampling and Blurring operators: Blurring was implemented as a Gaussian kernel with a standard deviation of 1 voxel. The blurred image was then down-sampled by averaging every 8 voxels (to simulate the partial volume effect), resulting in half of the original resolution. Update of the motion parameters: The motion parameters were estimated only once prior to the reconstruction and were kept fixed during the optimization. These parameters were used for other comparing methods as well. However, it would be of great interest for our future works to add more two-step cycles of motion estimation/reconstruction and follow the changes of the estimated parameters for convergence.