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

Authors

Lide Mu, Huafeng Liu

Abstract

Accurate reconstruction and imaging of cardiac transmembrane potential through body surface ECG signals can provide great help for the diagnosis of heart disease. In this paper, a cardiac transmembrane potential reconstruction method (GISTA-Net) based on graph convolutional neural network and itera-tive soft threshold algorithm is proposed. It fully combines the rigor of math-ematical derivation of traditional iterative threshold shrinkage algorithm and the powerful expression ability of deep learning method, as well as the char-acterization ability of graph convolutional neural network for non-Euclidean space data. We used this algorithm to simulate ectopic pacing data and simu-lated myocardial infarction data. The experimental results show that this al-gorithm can not only accurately locate the ectopic pacing point, but also ac-curately reconstruct the edge details of myocardial infarction scar while graph convolution makes full use of the connection information between the nodes on the heart surface.

Link to paper

DOI: https://doi.org/10.1007/978-3-030-87231-1_53

SharedIt: https://rdcu.be/cyhWc

Link to the code repository

N/A

Link to the dataset(s)

N/A

Reviews

Review #1

Please describe the contribution of the paper

The authors proposed GISTA-Net, a network that combines graph convolutional neural network and iterative soft threshold algorithm for cardiac transmembrane potential reconstruction. The authors used GISTA-Net to simulate ectopic pacing data and simulated myocardial infarction data. They conducted several experiments under different configurations to evaluate the proposed method. The results showed superior performance as compared to the state-of-the-arts and suggest GISTA-Net ability to reconstruct the edge details of myocardial infarction scar.
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.

I read the paper with a great interest The idea of integrating GCNN with ISTA is neat and provides better solution as it allows combining the strength of the traditional iterative threshold shrinkage algorithm with the powerful expression ability of deep learning method.

The methodology description is clear, and the formulation of GISTA-Net is novel and solid. The authors conducted several experiments under different configurations, and provided solid evaluation. The proposed method has strong clinical impact.
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 paper is very crowded. I suggest to remove some of the texts and replace it with figures or maybe write sentences concisely. Also, the paper has several grammar and punctuation errors (e.g., last sentence in the paper). Please do several rounds of proofreading before final submission.

The authors need to report the computational complexity of the proposed method as compared to ISTA and TV.  
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

It is hard to reproduce the paper based on the current description. I, however, think the authors did not provide details about the configuration and experiments due to the page limits. Providing the code would be great.
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 paper is very crowded. I suggest to remove some of the texts and replace it with figures or maybe write sentences concisely. Also, the paper has several grammar and punctuation errors (e.g., last sentence in the paper). Please do several rounds of proofreading before final submission.

The authors need to report the computational complexity of the proposed method as compared to ISTA and TV.  
Please state your overall opinion of the paper

Probably accept (7)
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

The idea of integrating GCNN with ISTA is neat and provides better solution as it allows combining the strength of the traditional iterative threshold shrinkage algorithm with the powerful expression ability of deep learning method.

The methodology description is clear, and the formulation of GISTA-Net is novel and solid. The authors conducted several experiments under different configurations, and provided solid evaluation. The proposed method has strong clinical impact.
What is the ranking of this paper in your review stack?

2
Number of papers in your stack

5
Reviewer confidence

Confident but not absolutely certain

Review #2

Please describe the contribution of the paper

This paper introduces a novel regularization approach for TMP estimation in ECGI. The main idea is to train a DNN that learns the residuals of the regularization model and then tries to minimize those in an L1 regularization framework.
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 this paper is the novelty of the method presented. It is a very interesting approach to introduce the prior knowledge learned with deep learning without directly imposing it during reconstruction. Its use of iterative soft-thresholding in this context is also quite good.
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.

Validation is the main weakness of this method. The paper shows many examples, but little details in each one. Did the authors include noise in their measurements before computing the inverse solutions? Otherwise the results will be overoptimistic. How were the simulations generated? What parameters were tuned?

More importantly, there are no details on how the network was trained. Specifically, it is not clear how the training and test examples were generated. Were multiple geometries used or just one? The ECGSIM heart models have ~1500 nodes, which means that the training dataset has considerably more examples than potential starting points of the PVC… that might imply that testing data was extremely similar to the training data.
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 of this paper is limited by the lack of details about the generation of the dataset and the training approach. These severely constrain the possible comparisons that could be done.
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
- Why did the authors use an adjacency matrix within a neural network to generate the forward model? Why not using the forward solutions computed with BEM (available in ECGSIM)? The paper also lacks detail about which graph was used to generate this adjacency matrix. Did it include torso and heart nodes? just heart?
- The paper also lacks details about the activation times estimation used. Based on the examples from Fig2, it is probably easy to estimate, but it is generally important to state how these were computed when PVC localization error is provided.
Please state your overall opinion of the paper

accept (8)
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

The novelty of the approach makes it considerably interesting. However, validation and experiments is flawed. The balance is in favor of accepting due to novelty.
What is the ranking of this paper in your review stack?

3
Number of papers in your stack

5
Reviewer confidence

Very confident

Review #3

Please describe the contribution of the paper

The author proposed to use GCN based iterative soft threshold network for Cardiac Transmembrane Potential Imaging.
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 problem is important and interesting as it tries to reconstruct signals from the cardiac surface. This approach can be applied to solving other inverse problems.
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 SNR level is still very high. The authors can test SNR = 0 dB or negative value. The motivation of using GCN is not clearly stated.
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

Confident about the reproducibility if the authors release code.
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
1. More experiments when SNR is much lower than the current ones.
2. Better explanation of motivation using GCN is appreciated.
Please state your overall opinion of the paper

Probably accept (7)
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

The novelty of method and interesting application make this paper interesting to read.
What is the ranking of this paper in your review stack?

2
Number of papers in your stack

5
Reviewer confidence

Very 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 proposed GISTA-Net, a network that combines graph convolutional neural network and iterative soft threshold algorithm for cardiac trans-membrane potential reconstruction. This is an intriguing, advanced application and all reviewers found the proposed method novel and solid. Weaknesses of the paper include (1) clarity of presentation (should be more concise), (2) insufficient validation and (3) missing implementation details. The authors are encouraged to complement the required information from the reviewers in the final version and submit code for reproducibility and dissemination of the knowledge.
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).

1

Author Feedback

N/A

back to top

Cardiac Transmembrane Potential Imaging with GCN Based Iterative Soft Threshold Network