Paper Info Reviews Meta-Review Author Feedback Post-rebuttal Meta-Reviews

Authors

Dong Wei, Kai Ma, Yefeng Zheng

Abstract

View planning for the acquisition of cardiac magnetic resonance imaging (CMR) requires acquaintance with the cardiac anatomy and remains a challenging task in clinical practice. Existing approaches to its automation relied either on an additional volumetric image not typically acquired in clinic routine, or on laborious manual annotations of cardiac structural landmarks. This work presents a clinic-compatible and annotation-free system for automatic CMR view planning. The system mines the spatial relationship—more specifically, locates and exploits the intersecting lines—between the source and target views, and trains deep networks to regress heatmaps defined by these intersecting lines. As the spatial relationship is self-contained in properly stored data, e.g., in the DICOM format, the need for manual annotation is eliminated. Then, a multi-view planning strategy is proposed to aggregate information from the predicted heatmaps for all the source views of a target view, for a globally optimal prescription. The multi-view aggregation mimics the similar strategy practiced by skilled human prescribers. Experimental results on 181 clinical CMR exams show that our system achieves superior accuracy to existing approaches including conventional atlas-based and newer deep learning based ones, in prescribing four standard CMR views. The mean angle difference and point-to-plane distance evaluated against the ground truth planes are 5.98 degrees and 3.48 mm, respectively.

Link to paper

DOI: https://doi.org/10.1007/978-3-030-87231-1_51

SharedIt: https://rdcu.be/cyhV8

Link to the code repository

https://github.com/wd111624/CMR_plan

Link to the dataset(s)

N/A

Reviews

Review #1

Please describe the contribution of the paper

This paper presents a novel framework for automated view planning from CMR images. In comparison to previous methods, the authors proposed to use the information contained on the DICOM images to find the intersection between the different views using a deep learning regression network. The outputs of these regression networks will be a heatmap that measures the distance to the intersecting line, which will then be aggregated to predict the localisation of the new view. Results of the paper demonstrate the high accuracy of the proposed method compared to other state-of-the-art methods.
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.
- Novelty of the idea and the proposed pipeline reduced the use of time-consuming annotations
- The paper is scientifically sound.
- Extend validation of the proposed method and comparison with other state-of-the-art methods
- Improved accuracy in estimation in results presented
- Potential clinical translation
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.

There is not a clear description of the deep learning networks used for each of the different parts of the pipeline. In the methods section, the authors describe the use of a regression network to estimate the heatmaps, but they don’t provide more details of the structure of the network. Then in the results section, the authors mention the use of a U-Net network, but it is not clear if they have replaced the standard U-net loss function with their loss function or how was trained the network. In the revised paper, I would suggest better explaining the use of the U-Net network for regression. From my point of view, the authors have proposed a new pipeline with strong clinical applicability and some rewriting of the paper will help to better support this.
Please rate the clarity and organization of this paper

Satisfactory
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 some lack of clarity on the explanation of deep learning networks, which makes it a little hard the reproduce the paper. However, the authors mention on the paper that they aim to release the source code after acceptance of the paper, which will help to make the paper more reproducible and that other researcher can use their proposed method.
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html
Major comments: Based on the title of the paper and one of the titles of the method sections “Target Plane Regression with SSL” it seems that they are using a self-supervised strategy. However, there is no clear definition of how they achieve this self-supervision more than they use the information contained in the DICOM images. Could please the authors provide a more detailed explanation on how they use SSL for their work? In the implementation section, the authors state that for the axial and pSA sequences, all the slices of a patient are treated as a mini-batch. However, the number of slices can vary between-subject. Could please the authors clarify how they handle this? Would the mini-batch size vary between subjects and only one subject is used per mini-batch, or what is the strategy? Figure 1 shows the input, the heatmap and then the optimal localisation of the intersecting line with the localiser, however, it does not show the generated images based on these planes. If possible, it would be interesting to see the predicted images based on the intersecting lines. Also, on the text and Fig S1, the protocol starts with the Axial view, but this is not reflected in Fig 1. I would suggest being better define on paper the different parts of the pipeline, and explain if the same method is used for the planning of the set of pseudo-view localisers and the standard planes. In clinical practice, the pSA is not always acquired as the localiser, p2C and p4C is sufficient to correctly plan the 2C, 3C, 4C and SAX sequences, how this will affect their proposed protocol? Minor comments:
- SSL is never defined but used in a title, maybe replace by self-supervised learning
- Figure 2, the text in white is split (or overlay) by the colour lines and sometimes is hard to read. I would suggest replacing the figure and place the text on the edges of the figure to make it easier to read
- No acknowledgement section, which will need to be inside of the limit of 8 pages.
- The paper starts with a sentence related to ischemic heart diseases; however, the proposed method is not limited to this group of patients. CMR imaging is prescribed for anyone with suspected heart disease (ischemic or not ischemic) and from my point of view, it will strengthen the paper to say that this method works for any sort of patients as CMR is the goal standard.
Please state your overall opinion of the paper

borderline accept (6)
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

This paper proposed a novel approach and I think is very interesting the way that the authors overcome the need of manual annotations from existing techniques, however, the description of the methods is not very clear and could be improved for clarity and reproducibility of the paper
What is the ranking of this paper in your review stack?

1
Number of papers in your stack

2
Reviewer confidence

Very confident

Review #2

Please describe the contribution of the paper

The authors propose a deep learning-based, self-supervised learning, framework for fully automatic view planning for Cardiac Magnetic Resonance (CMR) imaging. The proposed framework appears to be compatible with the busy clinical practice schedule.

The paper appears well written. The topic and the application are of interest and much needed for the busy clinical setting, where time is spent manually planning the scan before acquiring the images of interest.
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 architecture is fully automated and therefore does not require manual annotations, in comparison with previously published work, which require these manual annotations for the training. This is achieved by predicting the intersection lines between different views.

The proposed method is compatible with clinical practice, saving valuable time used otherwise in manually planning the scan.

The proposed work does not require the acquisition of a 3D volume in order to prescribe the standard CMR views.
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.

Perhaps the three chamber view (3C) is not so accurately predicted as the HLA or the VLA.
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

Based on the information provided by the authors upon submission the proposed work appears reproducible at this stage.
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html

Page 3 Methods (page top half): Usually the LV two-chamber view is called VLA in clinical practice. Similarly, the LV four-chamber is called HLA. The LV three-chamber view is called LVOT. I recommend the authors at least to mention this terminology at this point, to further emphasize the direct clinical application of the proposed work.

Subjects are likely to slightly move during the scan. This can cause the acquired planes to be slightly misaligned. Has this aspect been investigated in any sense?
Please state your overall opinion of the paper

strong accept (9)
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

I believe the proposed work is in line with the MICCAI areas of interest and has academic merit an innovation. Furthermore, the proposed work appears to be directly applicable to clinical practice.
What is the ranking of this paper in your review stack?

1
Number of papers in your stack

4
Reviewer confidence

Very confident

Review #3

Please describe the contribution of the paper

Combines self-supervised learning and multi-view planning for optimal plane prescription.
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 approach: self-supervised learning by leveraging metadata contained in DICOM tags
- Thorough study design
- Extensive set of experiments to assess performance
- Complete and concise description of methods
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.

None
Please rate the clarity and organization of this paper

Excellent
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
Work is reproducible since:
- Dataset details are presented in supplementary information
- Deep learning implementation details are presented
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html

The authors are strongly encouraged to convert this work into a journal submission. The inclusion of a radiologist survey would further strengthen the work.
Please state your overall opinion of the paper

strong accept (9)
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
- Novel approach
- Clear presentation of content
What is the ranking of this paper in your review stack?

2
Number of papers in your stack

4
Reviewer confidence

Somewhat confident

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 novel method to help CMR acquisition planning by estimating the location of standard planes using heatmaps. The proposed method uses standard data which includes implicit labels encoded in the plane orientation available from DICOM data, to train the model, with no extra annotation efforts needed. The results show high accuracy suitable for clinical utilization. All reviewers have agreed in the merit of the propsed approach, and there are some comments concerning clarity that should be addressable.
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).

2

Author Feedback

N/A

back to top

Training Automatic View Planner for Cardiac MR Imaging via Self-Supervision by Spatial Relationship between Views