Authors

Beomgu Kang, Hye-Young Heo, HyunWook Park

Abstract

Magnetization transfer contrast magnetic resonance fingerprinting (MTC-MRF) is a novel quantitative imaging technique that simultaneously measures several tissue parameters of semisolid macromolecule and free bulk water. In this study, we propose an Only-Train-Once MR fingerprinting (OTOM) framework that estimates the free bulk water and MTC tissue parameters from MR fingerprints regardless of MRF schedule, thereby avoiding time-consuming process such as generation of training dataset and network training according to each MRF schedule. A recurrent neural network is designed to cope with two types of variants of MRF schedules: 1) various lengths and 2) various patterns. Experiments on digital phantoms and in vivo data demonstrate that our approach can achieve accurate quantification for the water and MTC parameters with multiple MRF schedules. Moreover, the proposed method is in excellent agreement with the conventional deep learning and fitting methods. The flexible OTOM framework could be an efficient tissue quantification tool for various MRF protocols.

Link to paper

DOI: https://link.springer.com/chapter/10.1007/978-3-031-16446-0_37

SharedIt: https://rdcu.be/cVRTw

Link to the code repository

N/A

Link to the dataset(s)

N/A

Reviews

Review #1

Please describe the contribution of the paper
1. OTOM can be applied to any MRF schedules unlike the previous deep learning based studies dedicated to only a single fixed MRF schedule.
2. It enables transfer learning of the pre-trained OTOM on each dataset of new MRF schedules to further improve the accuracy while significantly reducing data preparation and network training 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.

An Only-Train-Once MR fingerprinting (OTOM) framework that estimates MTC tissue parameters from MR fingerprints regardless of the MRF schedule was proposed, thereby avoiding time-consuming process such as generation of training dataset and network training for different MRF schedules. The deep neural networks were constrained to a single MRF schedule corresponding to the training dataset. If the MRF schedule is changed, the deep neural network has to be trained with new training dataset that was generated with the new MRF schedule. This process is very time consuming and inefficient.So the utility of MRF techniques would benefit greatly from the development of streamlined deep learning frameworks or even only-train-once methods for various MRF sequences. And OTOM can be applied to any MRF schedules unlike the previous deep learning based studies dedicated to only a single fixed MRF schedule.
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.

It’s hard to reproduce, because they did not share their codes.
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

Negative for the reproducibility of the paper, because they did not share their codes.
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

It is difficult to reappear, but the content is very interesting.
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?

They proposed an Only-Train-Once MR fingerprinting (OTOM) framework to estimate water and MTC parameters from MR fingerprints regardless of the MRF sequence. Unlike the previous deep learning studies, the proposed method was trained with numerous patterns and lengths of MRF sequence, allowing them to plug any MRF sequence rather than a fixed MRF sequence. The flexible OTOM framework could be an efficient tissue quantification tool for various MRF protocols.
Number of papers in your stack

7
What is the ranking of this paper in your review stack?

7
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

Review #2

Please describe the contribution of the paper

The authors propose a model that can sustain changes in the MRF schedule and yet produce consistent tissue parametric maps.
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. Application of the bidirectional LSTM seems apt for the magnetization evolution signal
2. The ability to apply it to different MRF schedules is attractive although a clear explanation of how this happens needs to be detailed. Please see next section
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 study needs to incorporate explainable AI components to demonstrate the tasks that the bi-LSTM is actually learning.
2. The lack of a gold standard MT experiment and only comparing the MRF reconstruction methods is weak given that the MRF MT maps themselves need to be validated for clinical use. This needs to be at least included in discussion.
3. Random sampling of schedule components enables the generalization to different schedules but the use of TL to specific schedule contradicts this feature. So, the implementation can be a container model reference which would be optimal once a TL is employed. It is best to state this explicitly and underscore the benefit of a generalized model as a baseline rather than claim that the model can deal with any schedule. The reviewer recommends toning down this claim which might be confusing.
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 reproducibility metrics have been met
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 attempted an important problem in the field of saving training time and computational needs for MRF acquisition schedules. The approach of using a bi-LSTM seems appropriate. However, there are some improvements related to explainable AI and moderating of claims which might make this manuscript more impactful. These are outlined in the strengths and weaknesses sections.
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?
1. Problem is well motivated
2. LSTM approach seems appropriate
3. Claims need to be toned down - random sampling necessarily does not extrapolate to every scheme and especially needs further testing and validation for generalization. The use of TL indicates some of these challenges.
Number of papers in your stack

5
What is the ranking of this paper in your review stack?

4
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

Review #3

Please describe the contribution of the paper

The authors propose a recurrent neural network-based (RNN) approach that is able to estimate MTC tissue parameters from MR fingerprints acquired with various MRF schedules, avoiding the time-consuming process of generation of training dataset and network training for different MRF schedules. The proposed method shows similar accuracy to the conventional fully connected neural network (FCNN) method.
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 Only-Train-Once MR fingerprinting (OTOM) seems a promising solution of estimating quantitative parameters from different MRF schedules. Solid investigation of the OTOM method in digital phantom and in vivo 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 state-of-the-art baseline method of dictionary matching is not included.
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 is well described though the code is 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: The dictionary matching method should be included as one of the baseline methods. 2: The in vivo data is acquired with 4x acceleration. Any reconstruction to reduce the undersampling artifacts before the parametric mapping? 3: How was the Gaussian noise level determined? Please clarify.
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 Only-Train-Once MR fingerprinting (OTOM) seems a promising solution of estimating quantitative parameters from different MRF schedules, eliminating the process of lengthy data preparation and model retraining for a different MRF protocol.
Number of papers in your stack

5
What is the ranking of this paper in your review stack?

1
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.

The paper proposes a model to sustain changes in the MRF schedule and yet produce consistent tissue parametric maps without the necessity of additional training data. All three reviewers underlined the potential of the work and therefore I recommend acceptance.
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 would like to thank the reviewers for their thoughtful and constructive comments.

R2-1. Concerns about the lack of a clear explanation of how the proposed network works to support various MRF schedules. A) The mechanism of the proposed OTOM is based on the two-pool Bloch-McConnell equations. The final signal equation takes tissue and scan parameters as inputs and calculates MTC signal as an output. Therefore, three parameter spaces are interconnected: MR signal space, tissue parameter space, and scan parameter space. However, the scan parameter space is fixed in the conventional learning-based and dictionary-matching MRF techniques because the training dataset and dictionaries were simulated with numerous combinations of tissue parameters and specific scan parameters (single MRF schedule). Herein, various tissue parameters as well as scan parameters were considered in MRF reconstruction, allowing the neural network to fully explore the parameter space of two-pool Bloch-McConnell equations. This characteristic resembles that of Bloch fitting method but dramatically reduced computation time.

R2-2. The lack of a gold standard MT experiment and only comparing the MRF reconstruction methods is weak given that the MRF MT maps themselves need to be validated for clinical use. A) As the reviewer commented, this paper lacks comparison studies to a gold standard MT experiment. However, our proposed method cannot be systematically evaluated in vivo due to the lack of an objective ground-truth. A “true” gold standard does not currently exist for absolute MTC tissue parameters of in vivo brain tissue (Kim et al., NeuroImage. 2020 Nov;221: 117165). On the other hand, Kang et al. recently compared the MTC-MRF to the traditional steady-state MT experiment using synthetic MRI analysis and demonstrated that the deep-learning MTC-MRF had the overall lower errors for tissue quantification (Kang et al., MRM. 2021 Apr; 85(4):2040-2054). The difference in the acquisition schedule might have caused the performance difference. The synthetic MRI analysis will be needed to validate for clinical use in the future study.

R2-3. Random sampling of schedule components enables the generalization to different schedules but the use of TL to specific schedule contradicts this feature. So, the implementation can be a container model reference which would be optimal once a TL is employed. It is best to state this explicitly and underscore the benefit of a generalized model as a baseline rather than claim that the model can deal with any schedule. A) This is a very important comment. We agree that the use of TL (Transfer learning) to specific schedule contradicts the only-train-once feature. Therefore, we will revised the use of TL for validating the effectiveness of the OTOM in the manuscript. There were only small performance discrepancies between the OTOM and FCNN methods in the first place (Table 1). In addition, the in vivo tissue parameter maps obtained from the two methods were in excellent agreement (Fig. 4). Therefore, the performance gain from the use of TL is negligible, indicating the OTOM network is well optimized for each schedule. We will discuss this in the Result and Discussion section.

R3-1. The in vivo data is acquired with 4x acceleration. Any reconstruction to reduce the undersampling artifacts before the parametric mapping? A) The undersampling artifacts were corrected during reconstruction by minimizing the linear combination of the following loss terms: a data consistency term, a total variation, and a sparse wavelet regularization (Heo et al, Magn Reson Med. 2017 Jan; 77(2), 779–786).

R3-2. How was the Gaussian noise level determined? Please clarify. A) The Gaussian noise level was determined from the measured SNR of the acquired in vivo images to make the simulated environment reflect the realistic tissue properties. We will add this to the manuscript.

back to top

Only-Train-Once MR Fingerprinting for Magnetization Transfer Contrast Quantification