Paper Info Reviews Meta-review Author Feedback Post-Rebuttal Meta-reviews

Authors

Priscille de Dumast, Thomas Sanchez, Hélène Lajous, Meritxell Bach Cuadra

Abstract

Tuning the regularization hyperparameter α in inverse problems has been a longstanding problem. This is particularly true in the case of fetal brain magnetic resonance imaging, where an isotropic high-resolution volume is reconstructed from motion-corrupted low-resolution series of two-dimensional thick slices. Indeed, the lack of ground truth images makes challenging the adaptation of α to a given setting of interest in a quantitative manner. In this work, we propose a simulation-based approach to tune α for a given acquisition setting. We focus on the influence of the magnetic field strength and availability of input low-resolution images on the ill-posedness of the problem. Our results show that the optimal α, chosen as the one maximizing the similarity with the simulated reference image, significantly improves the super-resolution reconstruction accuracy compared to the generally adopted default regularization values, independently of the selected reconstruction pipeline. Qualitative validation on clinical data confirms the importance of tuning this parameter to the targeted clinical image setting. The simulated data and their reconstructions are available at https://zenodo.org/record/8123677.

Link to paper

DOI: https://doi.org/10.1007/978-3-031-43990-2_32

SharedIt: https://rdcu.be/dnwLN

Link to the code repository

N/A

Link to the dataset(s)

https://zenodo.org/record/8123677


Reviews

Review #1

  • Please describe the contribution of the paper

    The paper has proposed a simulation-based method to optimize the hyperparameter in solving the inverse problem for the reconstruction of super-resolution fetal brain MR image. The proposed method does not require constructing surrogate ground truth and provides a feasible quantitative framework for hyperparamter optimization.

  • 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.
    1. The paper presents a simple but efficient method using a fetal brain numerical phantom for quantitative optimization of regularization hyperparameter. This has addressed a long-standing problem in image recstruciton domain, particularly for fetal brain MR reconstruction.
    2. The paper devised an intelligent way to incorporate motion artifact within the simulation data to simulate the real-world clinical data.
    3. The paper has also outlined an important findings: system specific optimization of the regularization parameter is sufficient; subject specific optimization is not needed. This finding will be very helpful and will be inspiring for the clinicians/MR techs to know that it will be suffice to optimize the hyperparamter only once for each MR setting, not everytime as a new subject comes.
  • 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 clinical assessment of the reconstructed images were not provided; however, the authors mentioned this in their discussion.
  • 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 paper has provided enough detail to ensure reproducibility.

  • 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
    1. “In particular, we match the number and the orientation of the LR series, as well as the amplitude of fetal motion (from little to moderate), and the GA of each subject. “: how the fetal motion were estimated from the clinical images? Was this estimation qualitative? How these motion were simulated?

    2. “We observe that using the optimal parameters α∗1 and α∗2 makes the reconstructed images more similar”. It was surprising as the authors mention that matching the number of LR series is important to find the optimum hyperparameter. Were the number of LR series same for all 20 patients? If not, why α∗1 and α∗2 had similar effect?

    3. Table 2 shows statistical significance for SSIM and PSNR values even for those cases when the measures were same for both optimum and default regularization parameter.

  • 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?

    The proposed approach is uniqe and will have significant impact in MR image reconstruction.

  • 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 #2

  • Please describe the contribution of the paper

    The paper proposes a novel simulation-based approach for the regularization hyperparameter optimization, which plays a important role in ill-conditioned inverse problems. This method is evaluated in the fetal brain MRI.

  • 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.

    Generally, this paper is well-organized and the experimental setting is reasonable and the result shows its superiority.

  • 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 method seems only tested on the synthetic data. If this method can be tested on the real data, the result could be more convincing.

  • 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

    Yes

  • 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
    1. This work proposes the first approach to optimize the setting of the regularization parameter α based on numerical simulations of imaging sequences tailored to clinical ones. Not quite sure why such parameter α is important for the super-resolution task.

    2. Seems the work is highly based on Fetal Brain MR Acquisition Numerical phantom (FaBiAN), which could provide the LR HR pairs. But once we have the training pair, lots of supervised method can be applied, which reduces the importance of the proposed method.

    3. Like mentioned before, the method seems only tested on the synthetic data. If this method can be tested on the real data, the result could be more convincing.

  • 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

    4

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Like above

  • Reviewer confidence

    Not 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 #3

  • Please describe the contribution of the paper

    The authors proposed a simulation-based method for tuning the regularization hyperparameter alpha in fetal brain MRI super-resolution reconstruction. The proposed method was evaluated with both simulated and clinical 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 proposed method could be potentially effective and useful in clinic.

  • 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 clarity and organization of this paper can be further improved (see comments).

  • 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

    Nil

  • 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

    1.The simulation with rigid motion should be further elaborated. 2.Images of different orientations, especially those with relatively low resolution, are needed to directly demonstrate the performance with different hyper-parameters. 3.Results for data with substantial motion is also desired to demonstrate the robustness of the proposed method.

  • 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 proposed method is novel and practically useful. The clarity and organization of the paper should be further improved for readers to follow. More results from simulation/clinical study are desired to support their claims.

  • 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 paper describes a simulation-based method for the optimization of the hyper-parameter alpha in solving the inverse problem for the reconstruction of super-resolution fetal brain MR scans. The paper is well written and well organized. It has mostly appropriate experimental evaluation. Reviewer 2 was less knowledgable about the research area and topic. He/she acknowledged being less confident of his/her evaluation. Clarification questions were raised by the reviewers. We ask the authors to address them and revise the paper accordingly.




Author Feedback

This manuscript proposes a novel quantitative simulation-based method for optimizing the hyperparameter that balances data fidelity and regularization strength for enhanced image quality in fetal brain MRI super-resolution reconstruction (SRR). We are thankful to the reviewers and meta-reviewer for their feedback on our work. We have carefully accounted for their comments to increase the clarity of the camera-ready paper and would like to reassure them in this rebuttal about some legitimate concerns. Simulation of rigid motion (R#1, R#3). We have clarified that the amplitude of realistic fetal motion was qualitatively estimated from the clinical low-resolution (LR) series. As a complement, we rely on the three levels of fetal motion described in [1] and further documented by [2,3], with stochastic rigid motion simulated during k-space sampling to mimic as closely as possible the acquisition process. In this study, images were generated with little-to-moderate amplitude of 3D rigid motion, corresponding to a maximum of 5% corrupted slices over the whole fetal brain volume with independent translation within a uniform distribution of [-1,1]mm in every direction and 3D rotation within [-2,2]° for little motion, respectively [-3,3]mm and [-5,5]° for moderate motion. Tests on real clinical data (R#2). The very structure of our manuscript is built around two experiments: 1) controlled in silico environment, and 2) clinical environment. This organization seems to have convinced R#1 and R#3 of the feasibility of our approach, as well as of its significant impact on MR image reconstruction. However, let us recall that the lack for 3D isotropic high-resolution (HR) MR acquisitions of the fetal brain is a major hurdle for the development and validation of post-processing techniques. Therefore, besides the method for hyperparameter optimization – which could indeed, as R#2 pointed out, resort to the supervised training of a deep-learning network using the LR/HR pairs generated by our numerical framework –, this work also promotes the value of FaBiAN, a Fetal Brain MR Acquisition Numerical phantom [1], to provide highly realistic data that enable the quantitative evaluation of post-processing strategies where ground truth data are missing. Robustness to substantial motion (R#3). The generalizability and flexibility of our simulation environment allows to generate a broad variety of clinical scenarios, including the acquisition of LR series highly corrupted by motion, noise, as well as cases with very limited data available, along with the corresponding HR ground truth images, to further extend this work. However, such a systematic study would go beyond the scope of this paper due to space constraints. Similar effect observed on the SRR with α1* and α2* (R#1). As a support to Fig. 3, we have clarified that using the optimal regularization hyperparameter peculiar to every SRR pipeline instead of the default α reduces the domain shift between SRR techniques for a given subject (and not among all subjects which were indeed reconstructed using different numbers of LR series). Images of different orientations highlight the performance of the proposed strategy (R#3). For comparison, we have added a new figure (Supplementary Material Fig. 9) with clinical LR series in the three orthogonal orientations and corresponding SRR (MIALSRTK) using the default and optimized hyperparameters for a representative subject of 33 weeks of gestational age. Statistical significance (R#1). We acknowledge that Table 1 might be confusing as it reports the mean metrics computed across all subjects for the four configurations studied in Experiment 1, but also includes information about statistical significance from the paired Wilcoxon rank sum test based on the median of the different distributions. Yet, we decided to keep this information combined for the sake of space. [1] Lajous et al., 2022 [2] Rousseau et al., 2006 [3] Oubel et al., 2012



back to top