List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Xufeng Huang, Chengjin Yu, Huafeng Liu
Abstract
Recovering cardiac transmembrane potential (TMP) from
body surface potential (BSP) plays an important role in the noninvasive
diagnosis of heart diseases. However, most current solutions for TMP
recovery are typically proposed and designed to follow a static mapping
paradigm between TMP and BSP, which ignores the inherent dynamic
activation process of cardiomyocytes during the cardiac cycle. In this
paper, we propose to introduce the physiological information of this dy-
namic activation process in the objective functions. Based on this, we
further establish a physiological model based deep learning framework
for cardiac TMP recovery. First, the objective functions of our physio-
logical model are deduced via a two-variable diffusion-reaction system,
where the static mapping and the dynamic activation process of car-
diomyocytes are jointly modeled. Then, a data-driven Kalman Filtering
network (KFNet) is adopted to solve the proposed objective functions.
Specifically, the KFNet consists of two components: a state transfer net-
work (SSNet) is employed for directly predicting the prior estimation;
furthermore, a Kalman gain network (KGNet) is employed for adaptively
learning the gain coefficients. In our experiments, the proposed physio-
logical model is verified on the 1200 simulated subjects. The quantified
analysis shows the proposed method can accurately recover the TMP,
with the low LE values 10.5 for the ectopic pacing location task and
the high SSIM values 0.75 for the myocardial infarction detection task.
These powerful performances completely verify the effectiveness of our
model.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16434-7_42
SharedIt: https://rdcu.be/cVRsa
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 introduce a deep-learning method to perform inverse solutions in ECG. The network the authors use, emulates the general approach of a Kalman filter with the non-linear model provided by the deep-learning model. They test this method on synthetic data.
- 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.
Using dynamics to approach the inverse problem in ECG is indeed necessary and useful. That is particularly true for 3D (or 2D endo-epi) cardiac models. Adapting a Kalman filter to work for a non-linear dynamic model with neural networks is interesing as well.
- 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 limitation of the paper is validation. The authors provide little details about it. Despite the claim that they tested this method on 600 subjects, in reality ECGsim provides a maximum of 3 geometries, since the authors did not specify, the logical assumption is that they only used 1 and simulated multiple cases on that geometry. Failure to test on multiple geometries leads to over-optimistic results, specially for data-driven approaches. The authors did not specify whether they are adding noise to the measurements or not. In its absence, all the results will be over-optimistic. Separating experiments by location and ischemia is logical and acceptable, however, the reader is left wondering whether the network could not generalize to all cases. No information about which activation times or localization method is provided. Without it, it is not clear how generalizable are the results on the corresponding metrics.
- 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
Data can be reproduced from publicly available sources, although the specific data for training is not specified. Little detail is provided in the generation of the data.
Source code for the method is not provided.
Not enough information is provided with regards to validation.
- 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
This paper would benefit from further improvements in validation. The outstanding question that all inverse ECG papers should answer is how generalizable is this approach to different subjects. This is particularly sensitive for data-driven approaches that require training and testing separately.
More importantly, the authors should include noise of some form in the data. Otherwise, the results will always be good, but not realistic.
I would encourage exploring the ECGI database in EDGAR (https://edgar.sci.utah.edu/).
As a minor comment, it would be useful to clarify what are the axis in Figure 4(b)
- 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?
Despite the lacking validation, the method novelty is interesting and worth discussing.
- Number of papers in your stack
4
- 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
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Review #2
- Please describe the contribution of the paper
The paper presents a novel framework for predicting TMP from BSP using data-driven Kalman filtering network. The paper introduced a novel method and compares its results to various recent methods with favorable results for the presented 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 paper presents a novel framework for predicting TMP from BSP using data-driven Kalman filtering network. The paper introduced a novel method and compares its results to various recent methods with favorable results for the presented method.
- 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 results are produced on simulated data without discussing the size of simulated data (is it good enough size) and what would be the challenges faced with application of this method on real data. The paper also doesn’t have a thorough discussion, including any limitations of this work, in particular with respect to its application on real data or comparing to other existing recent methods.
- 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 paper lacks in sharing details about implementation and network architecture and it’s not open source. It uses a software to generate simulated data but generation parameters are not shared and the generated data is not shared. This will likely make it impossible to reproduce the results.
- 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
proofread and fix grammatical errors e.g. ‘is verify’ on 6th last line of abstract; ‘those powerful’ on 2nd last line of abstract; ‘a imprecise’ after eq 9; ‘a precise results’ after eq 9; ‘to instead the’ in conclusion;
proofread and fix typo/formatting errors e.g. text after eq 1, text after eq 4
fig 4 right part is not clearly legible
- 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?
interesting method results shown on limited size simulated data without a commentary about its application on real data
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
3
- 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
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Review #3
- Please describe the contribution of the paper
Authors propose an ECGI framework that combines deep learning with the physiological model of TMP dynamics. They argue that they address the limitation in the related works where they consider ECGI as an static mapping at each time step and ignore the temporal pattern of this problem. Considering the ECGI problem as state-space model they introduce a Kalman Filtering network consisting of a transition network and a kalman gain 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.
Designing a hybrid physiologically meaningful DL-based network for recovering TMP from BSP and embedding state update equations into the DL-based framework and which makes this approach interpretable.
Taking into account the inaccuracy in the state transition equation and adding the noise variable to compensate for that.
Interesting design of the loss function to ensure the accuracy of the state transition results and the final results.
Careful design of experiments and superior performance compared to a few related works.
- 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.
Authors argue the lack of literature in considering the temporal dynamics in this problem. However a recent work, Jiang et al 2021 titled “Label-Free Physics-Informed Image Sequence Reconstruction with Disentangled Spatial-Temporal Modeling”, consider a very similar approach to solve the ECGI problem. They similarly consider the state space model approach and learn the transition model and the temporal dynamics in the latent space of their model and use a decoder network as emission to the observation space. Authors need to discuss how their work is compared to this work.
Authors need to revise the text as there were a few grammatical and punctuation errors in the paper.
- 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
Authors will make the code 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
See above.
- 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?
Embedding the physiological model of TMP recovery inside the DL-based network makes this approach novel and interpretable. The only concern is that how this work is compared to a recent study with a similar approach.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
2
- Reviewer confidence
Somewhat Confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
6
- [Post rebuttal] Please justify your decision
Initially, I based the score on the 1)novelty of the approach 2) proposing a hybrid and physiologically meaningful framework 3) superior performance compared to some of the related works as the strengths of this work and 1)Missing the review of some related literature with similar approach to the same problem, as the weakness of this paper.
The author’s response to the reviews was almost acceptable since they 1) mentioned the difference of their approach to the pointed out missing reference and that they will expand their literature review. 2)Authors provided a few more results to support their method. They indicated that the method shows almost a stable performance in a more realistic setting when noise is included in the data. 3) they will revise the text to address some of the errors.
Even though the experiments could be expanded to include more geometries and clinical data, I believe this work has a strong methodological contribution and shows some promise. I will stick with my earlier decision of accepting this paper.
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 presents cardiac TMP recovery method based on deep learning via physiological model.
Key strengths 1) valuable application 2) reasonable methodology 3) good performance
Weakness 1) Some details are missing in the experiment. 2) Lack of sufficient about the challenge and limitation. 3) Some lack of literature.
Three reviewers have given the following comments: 1) R1 mainly considers the method is necessary, useful and interesting. 2) R2 mainly considers the method is novel, and the results is favorable. 3) R3 mainly considers the method is interesting and interpretable, and the performance is superior.
- 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
In our paper, ‘using dynamics to approach the inverse problem in ECG is indeed necessary and useful’(R1). Then ‘a novel framework for predicting TMP from BSP using data-driven Kalman filtering network’(R2), ‘which makes this approach interpretable.’ (R3). We thank the reviewers for praising our method as ‘necessary(R1), interesting(R2R3), novel(R2), reasonable(R3) and good performance(R2R3)’. Nevertheless, our work still needs further clarification. Validation: (1) We thank R1,R2 for bringing the professional issue on the verifying the model in different cardiac geometries/real data. As R1’s point this is an ‘outstanding question that all inverse ECG papers should answer’. However, it is also a common problem [10,12] to obtain a large number of cardiac geometries/real data due to the difficulties of constructing the individual cardiac geometries/labeling the ground truth. Therefore, our current experiments mainly concentrate on verifying the feasibility of the proposed method following the similar settings as [10,12]. Even so, we must thank R1 for providing ECGI database in EDGAR. Over the past few days, we have carefully studied the provided dataset. This seems promising to solve the problem of ‘how generalizable to different subjects’. We hope to further explore this database to better answer the generalization to different subjects in the following work. Thanks again.
(2) Following R1’s comments, we have done noise experiments, as well as the experiment on jointly predicting location and ischemia(R1). Our model shows stable performance in the two experiments. For example, when 15dB Gaussian noise is added to the input data, our model achieves CC 0.70±0.08 and SSIM 0.71±0.13, which outperforms the best compared method CC 0.68±0.10 and SSIM 0.69±0.13. When the model was trained and tested on both ectopic pacing and infarction, the LE reached 14.7±8.9 and the SSIM for infarction reached 0.72±0.06, exceeding the best compared method LE 18.3±10.9, SSIM for infarction 0.70±0.10. However, due to the limitation of article length and the rules of MICCAI, we may not put all the results into the final version. (3) Following R2’s comments, we have done experiments on the size of training data, which is from 100 to 2000 subjects in 50 intervals, and the effect tends to stabilize at 400 subjects. So our data is sufficient. Limitation: (1) Since the present formulation is in a personalized setting, we intend to extend this architecture to learn a geometry-invariant model that can be trained on multiple heart models and applied on a new subject(R2). Literature: (1) Following R2,R3’s suggestions, we review more literatures e.g., Jiang et al 2021, in considering the temporal dynamics. Those methods mainly consider the temporal dynamics in a latent space learned from BSP. We consider the physiological information of TMP via joint modeling the temporal dynamics in TMP itself and the forward mapping from TMP to BSP. Presentation: (1) We will share our code and data on GitHub after the paper is accepted (R1R2R3). (2) Figure 4(b) is a linear regression graph that describes the fitting of prediction and ground truth, in which horizontal and vertical axes represent mean value of real and predicted TMP, respectively(R1R2). Figure 4(c) shows the performance analysis on two metrics based on the box diagram(R2). (3) Activation time refers to the time when each node on the cardiac surface first reaches the maximum TMP value within a heartbeat(R1). The location error is the Euclidean distance between the ground truth pacemaker and the reconstructed pacemaker. The smaller the distance, the more accurate the location (R1). (4) At last, we will revise all the writing problems in the final version(R1R2R3).
Post-rebuttal Meta-Reviews
Meta-review # 1 (Primary)
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
I agree with the decision of all reviewers.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- 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).
1
Meta-review #2
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The paper proposes an ECGI framework that combines deep learning with the physiological model of TMP dynamic. All reviewers agree that the paper has value in application, methodology and performance. I believe the authors have addressed the questions on related works, experiments and reproducibility in rebuttal. Therefor, I recommend acceptance.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- 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).
2
Meta-review #3
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The authors have carefully addressed the major points in the rebuttal. Novel experiments with noise show rebustness of their method. Using further anatomical geometries will be addressed in future work, which I find acceptable. I vote for acceptance of the paper.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- 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).
4