Abstract

Assessment of cardiac function typically relies on the Left Ventricular Ejection Fraction (LVEF), i.e., the ratio between diastolic and systolic volumes. However, inconsistent LVEF values have been reported in many clinic situations. This study introduces a novel approach to quantify the cardiac function by analyzing the frequency patterns of the segmented Left Ventricle (LV) along the entire cardiac cycle in the four-chamber-image of echocardiography videos. After automatic segmentation of the left ventricle, the area is computed during a complete cycle and the obtained signal is transformed to the frequency space. A soft clustering of the spectrum magnitude was performed with 7.835 cases from the EchoNet-dynamic open database by applying spectral clustering with Euclidean distance and eigengap heuristics to obtain four dense groups. Once groups were set, the medoid of each was used as representant, and for a set of 99 test cases from a local collection with different underlying pathology, the magnitude distance to the medoid was replaced by the norm of the sum of vectors representing both the medoid and a particular case making an angle estimated from the dot product between the temporal signals obtained from the inverse Fourier transform of the spectrum phase of each and a constant magnitude. Results show the four clusters characterize different types of patterns, and while LVEF was usually spread within clusters and mixed up the clinic condition, the new indicator showed a narrow progression consistent with the particular pathology degree.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/3315_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: N/A

Link to the Code Repository

https://gitlab.com/acarrera4/cardiac_functoin_analysis/

Link to the Dataset(s)

https://unaledu-my.sharepoint.com/:u:/g/personal/proyectocardio890fm_bog_unal_edu_co/EacGQV-PWWRAllVAsNTAWRkBw-zgGKWk5lJVWBqFBCa0Og?e=5j1Zjm

BibTex

@InProceedings{Car_Characterizing_MICCAI2024,
        author = { Carrera-Pinzón, Andrés Felipe and Toro-Quitian, Leonard and Torres, Juan Camilo and Cerón, Alexander and Sarmiento, Wilsón and Mendez-Toro, Arnold and Cruz-Roa, Angel and Gutiérrez-Carvajal, R. E. and Órtiz-Davila, Carlos and González, Fabio and Romero, Eduardo and Iregui Guerrero, Marcela},
        title = { { Characterizing the left ventricular ultrasound dynamics in the frequency domain to estimate the cardiac function } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15001},
        month = {October},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The authors propose and study a fourier-based decomposition of LV area in AP4 echocardiography, using clustering in spectral magnitude to show there may be some separation by pathology in this cluster space.

  • 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.

    -Not an unreasonable approach to providing a richer signal of cardiac function than the very coarse LVEF.

  • 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.

    -spectral decomposition of cardiac signals is not novel. -there are no quantitative conclusions to the work. No further study of this potentially richer signal was performed, e.g. predicting EHR-noted pathology,

  • 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

    -What are the units in figure 3 on the right? -Footnote 8 (really, footnote 1) appears to break anonymization. -“Cluster 4 can allocated to others” - can be allocated -“Clusters 1 were split into low compromises” - unclear on the meaning of compromises here, as well as the plural clusters. -limited publication information (journal title e.g.) for [8] at least. -Are there actual Dice or other measures of the accuracy of the segmentation itself. The whole downstream analysis depends on them. Further, to what degree are the downstream clustering and analysis affected by small changes in the segmentation accuracy?

  • 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

    Strong Reject — must be rejected due to major flaws (1)

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

    Highly limited novelty and no clear conclusions. Interesting exploratory analysis but not material for this venue.

  • 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

    Weak Reject — could be rejected, dependent on rebuttal (3)

  • [Post rebuttal] Please justify your decision

    author-provided rebuttal, as well as other reviewer comments, make clear there are laudable elements of the work. However, the work still represents interesting exploratory analysis and is not appropriate for the venue.



Review #2

  • Please describe the contribution of the paper

    The authors propose a novel method to characterize cardiac function from 2D echo images, based on the Fourier transform of the left ventricular area time signal followed by some clustering. This is supposed to perform better than ejection fraction for assessing cardiac diseases.

  • 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 method that considers the full temporal evolution of the left ventricular area instead of simply using the left ventricular ejection fraction.
    • Application to a large 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.
    • It is not evident that performing (linear) Fourier transform before clustering is necessary, especially as the clustering technique is nonlinear. This should be investigated more thoroughly.
    • Sections 2.1 (especially Algorithm 1), 2.5 and 3 are very hard to follow. I cannot tell exactly what is demonstrated in the paper. The spectral analysis, evaluation method and results should be described more clearly.
  • Please rate the clarity and organization of this paper

    Poor

  • 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

    Please address the above mentioned main weaknesses.

    Minor comments:

    • Page 7: “3and 4” -> “3 and 4”
    • Page 7: “this Cluster” -> “this cluster”
  • 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?

    It is important to illustrate the importance of using the full temporal evolution of LV area and not just the EF. However, in its current state the paper is hard to follow; it is not clear why the method is built as it is, and what conclusions can be drawn.

  • 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 #3

  • Please describe the contribution of the paper

    The authors suggest a new method to extract biomarkers to evaluate the cardiac condition from ultrasound images in 4 chamber orientation. These biomarkers are extracted from the study of frequency patterns from a series of images covering the cardiac cycle (as easily obtained from US). The objective is to provide new parameters in addition or instead of the ejection fraction to detect and classify pathologies (with an impact on the heart).

  • 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 strenght of this paper is to propose an usable way to extract parameters from kinetic images, from ultrasound images, and then provides a different way to estimate the cardiac function. My opinion is that it could be in addition of the ejection fraction, to detect and define accurately the grade of a pathology. Indeed, it is well-known that certain cardiac pathologies get a normal ejection fraction. The method is dedicated to US images in 4 chamber view, but in my opinion could be extend to other type of imaging, such as MRI. We can noticed that the usable database is various, and considers a lot of different pathologies.

  • 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.

    Even if the approach could be considered as an alternative to the ejection fraction to evaluate the heart, in my opinion, we cannot dispense with ejection fraction. The suggested parameters must be considered as additionnal to classical one. Moreover, I suggest the authors to compare with other validated and well-known parameters extracted from kinetic images, such as the peak filling rate, or peak ejection rate, for example. Moreover, it is not clear if there is a real added value in the diagnosis of cardiac pathologies compared to ejection fraction, and in particular, the consitution of the 4 clusters is not clear (according to cardiac pathologies).

  • 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 does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    A part of the method is ever published.

  • 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
    • I suggest to the reviewer to reorient the objectives of the work, and define the suggested new biomarkers are additionnal to the evaluation of the ejection fraction, and not consider them replacing ejection fraction.
    • Does the preprocessing step is automatic ?
    • The constitution of the cluster must be more detailed.
    • Analysis in the frequency domain is maybe too short. Maybe an example could help to understand the approach.
    • The title, and the introduction of the work is not relevant. It must be clearly stated that the method is developped for US, and by default it is not a generic approach.
    • Some typo in the references (for example, the journal is not indicated in the reference 8)
  • 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 paper focus for an alternative way to evaluate the cardiac function, in place of ejection fraction. The methodolgy is interesting and original. But the results are not oriented in this way, and I am not convinced about the added value of the approach in a clinical point of view.

  • 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




Author Feedback

Below we summarize the main concerns of the reviewers and our answers to them. *Importance and novelty of the Fourier transform analysis of the echocardiography signal: Spectral decomposition of echocardiography signals is not novel, a main example being tissue Doppler imaging. However, these frequential analysis methods use exclusively the magnitude information of the Fourier transform. What is new in the present proposal is how phase information is integrated into the analysis and the importance of considering this at categorizing different pathologies. In this case, phase information helps us to establish when a particular frequency component occurs during the cardiac cycle and use this to compare the severity of different cardiac stages. The combination of magnitude and phase allow us to refine the notion of distance between cases pushing similar cases closer and different cases farther away

*Significance of the results and usefulness of the findings: The experimental results demonstrate the proposed approach is more effective than the LVEF measure in capturing temporal patterns that differentiate between different levels of cardiac compromise, in particular 100 real cases with different severity levels of cardiac disease established by the clinical condition after the Stevenson score (A, B or C define different levels of severity). However, this was not readily apparent in the included plots. An additional ANOVA analysis revealed significant differences among the distance established by the frequency descriptor herein proposed for different severities of cardiac pathologies.

*Comments about the presentation of the method and organization of the paper suggestions: -Preprocessing: Section 2.1 describes the preprocessing undertaken to ensure comparability between the two databases utilized in the article. The preprocessing was semi-automated, with the echocardiogram cone’s region of interest being manually identified, while the remaining preprocessing stages were automated. This section will be clarified by adding this information.

-Segmentation: Segmentation accuracy on the clustering results, a great advantage of making the analysis in the frequency domain is that the accuracy of the segmentation is less important since the possibility these errors are accumulated during the whole cycle is negligible. In fact the temporal patterns of the ventricular area reflect not only the outcome ventricular capacity but rather the compensatory mechanisms which eventually could make simple volume relations look normal. Finally, measure of the Dice was included in the methodology, specifically in the sentence “In a previous work, this network demonstrated outstanding segmentation results (Dice of 0.93), …”

-Frequency Analisys: Algorithm 1 describes better in terms of the problem the proposed methodology, as follows Input: Temporal series of cardiac cases Output: Assigned clusters with updated magnitude using phase information Step 1: Mapping to the Fourier space For Each case in the dataset Apply Fast Fourier Transform (FFT) to obtain magnitude and phase spectrum End For Step 2: Finding out the frequential representant of the cardiac cycle Perform spectral clustering of the magnitude with 7,835 cases from the Echonet database and 99 cases from a local collection with registered clinic state Step 3: Re-estimating intra-cluster distances using phase information for those cases with known clinical history For Each of the 100 cases: Set case_i to a particular group Apply inverse Fourier Transform using the phase spectrum and a unitary magnitude for both case_i and the medoid Compute dot product of the reconstructed temporal series between case_i and the group medoid Recover angle (theta) = \arccos ( <medoid, case_i> / (|medoid| |case_i|) ) Set a new case_i distance as the norm of the sum of two vectors, one with norm 1 and the other 1 + (stored distance), and making angle (theta) Store this new distance for case_i End For




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’

    R4 does not provide strong arguments for a strong reject and increased rating to weak reject after rebuttal. The other reviewers recommend accept so the paper should be accepted.

  • 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).

    R4 does not provide strong arguments for a strong reject and increased rating to weak reject after rebuttal. The other reviewers recommend accept so the paper should be accepted.



Meta-review #2

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

    Reject

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

    After considering the reviewer’s feedback and after looking into the paper myself, I agree to #R4. The work has value but presents some initial analysis which need to be deepened. For me, the introductory part is very weak. It only refers to cardiac segmentation and LVEF estimation. However, there is a plethoria of work in the field of cardiac motion assessment and myocardial dynamics. Furthermore, cardiac motion has been previously described by fourier transform or wavelets and this was not mentioned in the paper,e.g. quick search revealed: https://ieeexplore.ieee.org/document/731978 or 10.21203/rs.3.rs-303334/v1 Instead of clustering I don’t fully understand why the authors have not performed a classification or show that the obtained values for the groups differ in a statically significant way. #R1 further mentioned that the presentation is poor. No comparison to other existing methods in cardiac motion analysis have been performed.
    I think the paper should not be presented at this venue.

  • 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).

    After considering the reviewer’s feedback and after looking into the paper myself, I agree to #R4. The work has value but presents some initial analysis which need to be deepened. For me, the introductory part is very weak. It only refers to cardiac segmentation and LVEF estimation. However, there is a plethoria of work in the field of cardiac motion assessment and myocardial dynamics. Furthermore, cardiac motion has been previously described by fourier transform or wavelets and this was not mentioned in the paper,e.g. quick search revealed: https://ieeexplore.ieee.org/document/731978 or 10.21203/rs.3.rs-303334/v1 Instead of clustering I don’t fully understand why the authors have not performed a classification or show that the obtained values for the groups differ in a statically significant way. #R1 further mentioned that the presentation is poor. No comparison to other existing methods in cardiac motion analysis have been performed.
    I think the paper should not be presented at this venue.



Meta-review #3

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

    Reject

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

    This paper is poorly written. The main contribution in this paper is Section 2.3 which only has nine lines, and the innovation is very limited. It is obvious that this paper is way below the MICCAI standard.

  • 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).

    This paper is poorly written. The main contribution in this paper is Section 2.3 which only has nine lines, and the innovation is very limited. It is obvious that this paper is way below the MICCAI standard.



Meta-review #4

  • 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’

    The paper received mixed reviews and the criticism relates to the limited novelty in the approach and the early stage of the work. The primary and secondary ACs disagreed in their final recommendation. This meta reviewer argues that the paper makes a valuable contribution despite its limitations. In particular, it is important to enhance the interpretation of cardiac data from ultrasound imaging and the paper indicates that image-based descriptors of cardiac ejection fraction add meaningful clinical information that can be intuitively integrated into care. The authors should improve the clarity of the presentation and highlight limitations in their discussion.

  • 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).

    The paper received mixed reviews and the criticism relates to the limited novelty in the approach and the early stage of the work. The primary and secondary ACs disagreed in their final recommendation. This meta reviewer argues that the paper makes a valuable contribution despite its limitations. In particular, it is important to enhance the interpretation of cardiac data from ultrasound imaging and the paper indicates that image-based descriptors of cardiac ejection fraction add meaningful clinical information that can be intuitively integrated into care. The authors should improve the clarity of the presentation and highlight limitations in their discussion.



back to top