Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

During a fetal ultrasound scan, a sonographer will zoom in and zoom out as they attempt to get clearer images of the anatomical structures of interest. This paper explores how to use this zoom information which is an under-utilised piece of information that is extractable from fetal ultrasound images. We explore associating zooming patterns to specific structures. The presence of such patterns would indicate that each individual anatomical structure has a unique signature associated with it, thereby allowing for classification of fetal ultrasound clips without directly feeding the actual fetal ultrasound content into a convolutional neural network.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/1820_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: N/A

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Als_Zoom_MICCAI2024,
        author = { Alsharid, Mohammad and Yasrab, Robail and Drukker, Lior and Papageorghiou, Aris T. and Noble, J. Alison},
        title = { { Zoom Pattern Signatures for Fetal Ultrasound Structures } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15004},
        month = {October},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    The paper explore associating zooming patterns to specific structures. Authors try to distinguish different fetal US frame by using Reverse Quasi-Zoom (RQZ) value which could potentially be time and resource efficient in low-resource and low-compute settings.

  • 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 zooming information sequence to distinguish different ultrasound views is a practical idea, and has clinical application value.
  • 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 experiments are too simple.
    • The writting is not good.
    • The image quality of the article is poor. especially for Figure 3, It contains little information but occupies a significant amount of space.
  • 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 does not provide sufficient information for reproducibility.

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

    No

  • 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

    The idea of the article is good, but both the method and experimental part are insufficient. The writing also needs to improve. More questions and comments are listed below.

    1. How much time does it take to predict the RQZ value, and what’s the overall efficiency of the method? As you clained that the proposed method is time and resource efficiency, you should add experimentals to prove these.
    2. What’s “the fine-tuning phase” refer to? For me, fine-tuning refers to the process of taking a pre-trained model and further training it on a specific task or dataset to improve its performance for that task. This seems to be inconsistent with what you expressed.
    3. The model was trained for 1000 epochs only with no more than 121 squences. I think the model might be overfitting. Besides, the whole paper did not have any comparison or ablation study which is unacceptable.
    4. What the meanings of Fig.5? It doesn’t get mentioned anywhere in the paper.
    5. Typos: for example in section 3 “In this work, we look at the target frames, and the 300 frames before each target frame, this covers the 90 seconds of finetuning before a standard frame”, 90 seconds should be 10 seconds.
  • 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?

    bad writing and simple experiments.

  • 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

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #2

  • Please describe the contribution of the paper

    The paper explores utilizing the zoom information in fetal ultrasound. The paper reveals that the zooming patterns is specific to different anatomical structures, which can act as ‘shortcuts’ for classifying fetal ultrasound clips.

  • 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.

    To my best knowledge, studying the zoom pattern in fetal ultrasound is very novel and meaningful, as it correlates to the standard plane definition from ISUOG (the anatomical structures need to be zoomed in to occupy e.g. half of the image to format a standard plane). The motivation for utilizing the zoom information for the low-resource fetal ultrasound classification is realistic and strong.

  • 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.

    More proofs (preferably from the clinicians) are needed to demonstrate the discovered correlation between zooming information and anatomical structures is not spurious, i.e., bias or shortcuts that harm the generalizability of the model.

  • 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 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?

    The method was validated in a public dataset. However, the code might need to be made public for reproducibility upon acceptance.

  • 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
    1. The supplementary material violates the rule for the appendix (https://conferences.miccai.org/2024/en/PAPER-SUBMISSION-AND-REBUTTAL-GUIDELINES.html) and was thus not reviewed. The lack of preliminary knowledge weakens the soundness of the proposed method. For instance, I do not understand how the Reverse Quasi Zoom values were computed.
    2. The writing of the paper could be improved.
    3. As mentioned in the weakness, the correlation of anatomical structures and the zoom information can be spurious (e.g., valid only for some specific operators or devices). Could you figure out why they are correlated based on the mechanism behind them?
    4. Why does the developed 1-D CNN only have 2 output channels? Should it be 3 (background, CRL, NT)?
  • 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 Reject — could be rejected, dependent on rebuttal (3)

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

    I do like this paper. My biggest concern lies in the generalizability of the discovered relationship between the zooming information and anatomical structures. I am happy to improve the score if I see more valid proof.

  • 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

    Accept — should be accepted, independent of rebuttal (5)

  • [Post rebuttal] Please justify your decision

    I would like to thank the authors for the rebuttal. The rebuttal addresses most of my concerns. For the supplementary material guidelines, I actually mean that the appendix should not exceed 2 pages (please carefully read the rules). The authors might want to be more careful on the rules in their future works.

    Further studies on the relationship between the zoom pattern signatures and the other metrics like devices and GA is expected, I can see the authors have promised this.

    In all, the clinicians’ view mentioned in the rebuttal addresses my biggest concern, and I am happy to improve my grade - the paper has reached the bar of MICCAI acceptance, in my view.



Review #3

  • Please describe the contribution of the paper

    This work introduces a novel approach to predicting anatomical classes using the zoom function, without relying on access to image content. This method represents a significant advancement in fetal screening techniques and adds to the arsenal of tools available to clinicians.

  • 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 work lies in its thoughtful approach and novelty within the field of fetal medicine. It introduces innovative methods for fetal screening, particularly in utilizing the zoom function for predicting anatomical classes without the use of the image content. This study represents original research of utilizing the ignored information in the ultrasound images and is the first of its kind. It has huge potential in training the sonographers to decide the zoom settings, concerning the poor quality of the fetal ultrasound images. I also see the potential of using such approaches in the countries where the screening of certain fetal planes are restricted.

  • 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.

    Predicting the anatomical class solely by relying on the zoom function, without access to the image content, can be challenging, as it may depend on the experience of the sonographer, potentially introducing bias. Apart from this I don’t find any other weakness in this paper.

  • 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?

    It would be beneficial to provide the dataset used in this work. Currently, it mentions the PULSE dataset, which may lead to confusion as it also refers to another dataset in signal processing. Therefore, it might be a good idea to provide a clear reference to the original dataset to facilitate the reproduction of this work.

  • 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

    It would be beneficial to mention a standard protocol for zooming different organs that is universally adopted across countries. Additionally, it would be interesting to conduct an experiment on the impact of variation in zoom settings for various patients and gestational ages, as zoom settings can vary based on patient habitus. Despite the relatively low reported accuracy, this work is well-thought-out and novel in the field of fetal medicine, opening many doors to innovative approaches in fetal screening.

  • 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

    Accept — should be accepted, independent of rebuttal (5)

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

    Please accept this manuscript, this is a novel approach in the fetal screening. Please consider to add an experiment with imapct of zoom based patient BMI, and fetal presentation.

  • 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

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Author Feedback

Thank you for recognizing our novel use of zoom.

R1.3. Our work involves 6 sonographers, reducing sonographer bias.

R1.7. We will explore this approach with other gestational ages in the future.

R4.3. Thank you for your comments. Firstly, we would like to note that two of the authors are clinicians who perform these scans on a daily basis. We strongly believe it is important for clinicians to be involved in such work. Secondly, we believe that the results are not spurious, primarily from the fact that intuitively the results make strong clinical sense and are clinically explainable. The CRL measurement requires the entire fetus to be visible on the screen before a measurement is made (therefore zoomed out), while NT requires a specific part behind the fetus’s head to be focused on for a measurement (therefore zoomed in to fill 75% of the screen according to the guidelines). At this age, the CRL is roughly 65 ± 19mm, while NT is usually around 1.1 to 3.0 mm, so to view them well, one would expect different zoom levels for each.

R4.7.1. The supplementary complies with section 5’s rule on “concurrent submission to MICCAI or another conference” and section 3’s PDF stating concurrent submissions are allowed if anonymized. RQZ values are determined by reading a few pixel lines in the depth scale region of a raw US image.

R4.7.3. In the supplementary, we show that models can differentiate zoom levels and are not fooled by post-acquisition manipulations like cropping or resizing. This was crucial for our current work. All US machines require fine-tuning and zooming for the correct view. Our work involves six operators, reducing operator bias.

R4.7.4. It has two output channels, one for CRL and one for NT. As shown in Fig. 2, frames frozen for interpretation or measurement have a class label. We predict the structure that the fine tuning phase clip is approaching by the zoom pattern. We do not predict the contents of the individual frames. A sonographer would approach and then freeze for a CRL or NT view for detection and measurement, not the background.

R3.3.1. Our experiments show that zoom sequences can be attributed to specific anatomical structures. We prioritize simplicity for low-compute settings and have effectively addressed our research question.

R3.3.3. Figure 3 depicts the confusion matrix, which we will reformat for clarity.

R3.7.1. RQZ values are inexpensive to obtain, relying only on reading specific pixel values in a raw US image. We can add details, as covered in the supplementary material. The proposed task operates on a different scale. Classifying a 300-integer sequence (0-4) with a 1-D CNN (272,770 parameters) is much cheaper than classifying a 300-frame video (224x224 pixels) with ViViT-B (88.9 million parameters). On a Dell i5 laptop with 8 GB RAM, e.g., the 1-D CNN takes 10 milliseconds per clip, while the video classifier takes 10 seconds.

R3.7.2. The fine-tuning phase involves a sonographer adjusting the probe to get the correct view, as described in Fig. 2’s caption. We classify the zoom pattern during this phase.

R3.7.3. This novel low-compute method avoids spatio-temporal data (US video clips) and uses zoom values instead. No comparable work uses zoom in this manner, so there’s no suitable comparison. Our model, a 1-D CNN with two convolutional blocks, is too small to ablate but addresses our main question: ‘Can we attribute zooming patterns to specific fetal structures?’ The difference between the test accuracy and the train accuracy is only 10%, indicating that the model generalizes to the test set and overfitting is not an issue. No data leaks to the test set, and early stopping is used to prevent overfitting.

R3.7.4. Fig. 5, referenced in the first Results and Discussion paragraph, is similar to Fig. 4. Fig. 4 uses the mean on the y-axis, while Fig. 5 uses the mode, as mentioned in the Fig. 5’s caption. Both y-axes are labeled.

R3.7.5. We’ll fix the typo.



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’

    N/A

  • 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).

    N/A



Meta-review #2

  • 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’

    This work should be accepted as the incorporation of clinical investigation of zooming patterns is very interesting and will generate significant discussion during the meeting among the community developing ultrasound-based CAI and MIC methods. The authors should improve the camera-ready version by including all the missing information that was provided in their rebuttal.

  • 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 work should be accepted as the incorporation of clinical investigation of zooming patterns is very interesting and will generate significant discussion during the meeting among the community developing ultrasound-based CAI and MIC methods. The authors should improve the camera-ready version by including all the missing information that was provided in their rebuttal.



back to top