Abstract

The Magnetic Resonance Fingerprinting (MRF) approach aims to estimate multiple MR or physiological parameters simultaneously with a single fast acquisition sequence. Most of the MRF studies proposed so far have used simple MR sequence types to measure relaxation times (T1, T2). In that case, deep learning algorithms have been successfully used to speed up the reconstruction process. In theory, the MRF concept could be used with a variety of other MR sequence types and should be able to provide more information about the tissue microstructures. Yet, increasing the complexity of the numerical models often leads to prohibited simulation times, and estimating multiple parameters from one sequence implies new dictionary dimensions whose sizes become too large for standard computers and DL architectures. In this paper, we propose to analyze the MRF signal coming from a complex balanced Steady-State Free Precession (bSSFP) type sequence to simultaneously estimate relaxometry maps (T1, T2), Field maps (B1, B0) as well as microvascular properties such as the local Cerebral Blood Volume (CBV) or the averaged vessel Radius (R). To bypass the curse of dimensionality, we propose an efficient way to simulate the MR signal coming from numerical voxels containing realistic microvascular networks as well as a Bidirectional Long Short-Term Memory network that replaces the matching process. On top of standard MRF maps, our results on 3 human volunteers suggest that our approach can quickly produce high-quality quantitative maps of microvascular parameters that are otherwise obtained using longer dedicated sequences and intravenous injection of a contrast agent. This approach could be used for the management of multiple pathologies and could be tuned to provide other types of microstructural information.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2226_paper.pdf

SharedIt Link: https://rdcu.be/dV1Oi

SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72069-7_25

Supplementary Material: https://papers.miccai.org/miccai-2024/supp/2226_supp.pdf

Link to the Code Repository

https://github.com/nifm-gin/MARVEL

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Bar_MARVEL_MICCAI2024,
        author = { Barrier, Antoine and Coudert, Thomas and Delphin, Aurélien and Lemasson, Benjamin and Christen, Thomas},
        title = { { MARVEL: MR Fingerprinting with Additional micRoVascular Estimates using bidirectional LSTMs } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15002},
        month = {October},
        page = {259 -- 269}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The main contributions of the paper are 1) to simulate magnetic resonance fingerprinting (MRF) signals in an efficient way, and 2) to estimate the quantitative parameters using bidirectional LSTM network.

  • 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 are 1) to estimate vascular microstructures with MRF sequence, which has the potential to be used for pathological cases and 2) to replace the matching algorithm with a bidriectional LSTM, which can improve the quality of the reconstructed maps as well as computational time.

  • 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 main weakness of the paper is the validation of the estimated cerebral blood volume (CBV) and averaged vessel radius (R), which seemed to noisy and relatively low-quality compared to other quantitative maps such as T1 and T2.

  • 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 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?

    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
    • There are differences between the reconstructed quantitative maps from dictionary matching and bidrectional LSTM, especially T1, T2, and R. Based on Table 1, there are more than 10% difference in WM of T1, GM in T2, and WM in R. It is not very clear how the best values were chosen or compared with other method or literature. The authors might need to provide supporting results to demonstrate the accuracy of bidirectional LSTM compared to dictionary matching.
    • Though bidirectional LSTM showed less noisy results, some of the maps seemed to be oversmoothed that it roused questions about the accuracy of CBV and R. For instance, the reconstructed CBV and R seemed to have more contrast between WM and GM regions though they are noisier than bidirectional LSTM.
    • To reduce the computational resources of the dictionary matching, it might be helpful to reduce the number of unknown parameters to be estimated. For example, B1 maps can be acquired separately with short scan time before or after MRF sequence, which can reduce the number of parameters and might improve the reconstruction time and SNR.
  • 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 Reject — could be rejected, dependent on rebuttal (3)

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

    Based on the comments mentiioned above especially for the main weaknesses, I would recommend weak reject for this paper.

  • 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 #2

  • Please describe the contribution of the paper

    This study propose an efficient method to simulate the MR signal coming from numerical voxels containing realistic microvascular networks. The method stores only a small signal dictionary and convolves it with the voxel’s frequency distribution to obtain the MR signal of that voxel. With the simulated dictionary, a Bidirectional Long Short-Term Memory (BiLSTM) network was trained to reconstruct quantitative parameter maps in a reasonable 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.

    During the construction of the simulated signal dictionary, this study only builds a standard dictionary containing signals with varying T1, T2, B1, and δf values. Subsequently, based on the δf distribution within each voxel, dictionaries for six parameters (T1, T2, B1, δf, CBV, and R) are obtained. This method circumvents the explosive growth of dictionary size in high dimensions, as it requires only a small standard dictionary to efficiently acquire the signal of any simulated voxel within a reasonable timeframe. By training a BiLSTM reconstruction network on a simulated dataset, this study circumvents the significant time required for dictionary matching, enabling more accurate and rapid reconstruction of quantitative parameter maps. The model’s reconstruction performance was validated on a real dataset.

  • 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 reconstruction results of different models are evaluated based on visual assessment of reconstructed images and the disparity between statistical distributions of tissue-specific metrics and theoretical distributions. However, quantifying evaluations are missing.

  • 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 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?

    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
    1. The authors are advised to provide a quantitative comparison of the reconstruction results of different models, which can be achieved through testing on simulated datasets.
    2. It is essential to provide detailed explanations in the manuscript regarding how the realistic vascular networks inside 3D voxels are obtained.
  • 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?

    The method proposed in this study demonstrates a well-structured approach, effectively addressing the drawbacks of large dictionary size and slow matching inherent in multi-dimensional MRF. However, further improvements are necessary to provide more comprehensive evidence of the model’s effectiveness.

  • 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 #3

  • Please describe the contribution of the paper

    The paper found an efficient way to simulate MRF signals coming from numerical voxels that have realistic microvascular networks. The authors believe that they only need to store a small number of dictionaries and simulations can be made fast on the demand. In addition, they used Bi-LSTM as their model since this model is great for sequential data which is suitable for MRF temporal sequence.

  • 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 greatly shows the depth of the knowledge of authors in this area and they have provided every detail in the implementation and mathematical equations. They used 3 healthy volunteer data as the data for their model which is real data and great. For the evaluation, they have a good comparison with other works and methods and their result shows a great improvement. Furthermore, the authors have suggestions regarding improving the result and know their faults or deficiencies.

  • 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 general concept and understanding of the paper are not very well and in some parts, the authors paid too much attention to details that may make it hard to follow the main goal of the paper.

  • 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 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?

    The authors provided the code in the supplementary and a lot of details for the parameters. Only the dataset which includes the 3 volunteers may be an obstacle for reproducibility.

  • 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

    The author can follow their findings on future work and complete them like improving network structure for longer fingerprints and realistic and diverse simulations.

  • 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?

    The overall score for this paper comes from the details the authors have put in the implementation and basic ideas and their idea for fast simulation which is needed in medical applications. In addition, they have used real data from volunteers which is very great for the medical society.

  • Reviewer confidence

    Somewhat confident (2)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • [Post rebuttal] Please justify your decision

    My concerns are addressed by the authors.




Author Feedback

We thank the reviewers for their insightful comments and for highlighting the main contributions of our paper, which include: 1) Simulating MRF signals depending, among other, on microvascular tissue properties (CBV and R), in an “efficient way” (R1, R3, R4). This efficiency stems from the fact that dictionaries do not need to be entirely stored. 2) Estimating 6 quantitative parameters within a “reasonable time” (R1) using a “suitable” (R4) BiLSTM neural network.

There were concerns raised by Reviewers 1 and 3 regarding the use of quantifiers to evaluate our method.

We fully acknowledge the need for additional validations of our method as pointed in our discussion.

There are several results that could be easily added to the paper such as: 1) an in-silico study of the dictionaries to assess the sensitivity of the MRF sequence to the different parameters; 2) Correlation plots between the results of the different models and those obtained using dictionary matching. All these plots are already available.

However, concerning the validation of our quantitative vascular estimates: our main objective was to show a proof a principle in human volunteers. In this case, obtaining ground truth values is difficult because of the injection of Gadolinium. Given the encouraging results obtained here, we will now start a new validation study in tumor patients that will receive contrast injection as part as their clinical exam.

While Reviewer 1 commended the clarity and organization of the paper, Reviewer 4 expressed concern that our attention to details might reduce understanding of the key messages.

We will consider emphasizing overarching concepts in the final version, yet we think that technical details are essential 1) For reproducibility (as pointed out by R4); 2) Due to the complex MRF setting considered in this paper, with the inclusion of microvascular parameters, necessitating a 2-step dictionary generation process. Regarding this specific vascular dictionary procedure, Reviewer 1 requested additional explanations on the procurement of realistic vascular 3D voxels. Briefly, we use public datasets of whole-brain healthy mice (due to unavailability of corresponding human datasets) microscopy. A segmentation pipeline applied on this data allows us to obtain a binary representation of vascular networks with realistic geometries. The CBV and R in each of the produced voxels are characterized and used as parameter dimensions in the dictionary. The complete process is detailed and discussed in an article, that will be published soon, as well as in past conference abstracts, which will be referenced in the final version.

Furthermore, Reviewer 3 raised concerns regarding the over-smoothing of CBV and R maps reconstructed with our BiLSTM.

We wanted to point out that the WM/GM ratio of 2 obtained in the CBV maps using the BiLSTM is actually expected in the human brain. It is however possible that the neural network reconstructions smooth the parameter maps, and we believe that additional factors can contribute to this issue: 1) Insufficient sensitivity of current MRF sequences to vascular parameters, motivating the design of new sequences; 2) Use of non-optimal vascular vessel geometries for simulating our signals. As explained in Section 4, we hope to replace our 28,000 structures by “more realistic and diverse vascular vessel geometries”. We aim to improve those points in future works.

Lastly, Reviewer 3 suggested acquiring the B1 parameter separately to speed up dictionary matchings. Although we attempted to use pre-acquired B1 maps to enhance matching time and BiLSTM reconstruction quality, we encountered two challenges that we definitely aim to tackle in future work: 1) Acquired B1 maps contain artefacts, which adversely affects reconstructions in the associated zones; 2) Non-negligible slice profile effects were observed with a fixed B1, despite efforts to mitigate these effects in our signal simulations.




Meta-Review

Meta-review #1

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    Two out of three reviewers have recommended accepting the paper, with the third giving a borderline reject score (not updated post rebuttal). The rebuttal response appears to address most of the major concerns raised and has acknowledged limitations of the study, such as the need for validation with larger datasets. At the outset, the method compares favorably in terms of quality, time and storage, though quantitative measurements in addition to Fig. 2 and 3 are currently lacking.

  • What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/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).

    Two out of three reviewers have recommended accepting the paper, with the third giving a borderline reject score (not updated post rebuttal). The rebuttal response appears to address most of the major concerns raised and has acknowledged limitations of the study, such as the need for validation with larger datasets. At the outset, the method compares favorably in terms of quality, time and storage, though quantitative measurements in addition to Fig. 2 and 3 are currently lacking.



Meta-review #2

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    N/A

  • What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/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).

    N/A



back to top