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

Abstract

Accurate fetal birth weight (FBW) estimation is essential for optimizing delivery decisions and reducing perinatal mortality. However, clinical methods for FBW estimation are inefficient, operator-dependent, and challenging to apply in cases of complex fetal anatomy. Existing deep learning methods are based on 2D standard ultrasound (US) images or videos that lack spatial information, limiting their prediction accuracy. In this study, we propose the first method for directly estimating FBW from 3D fetal US volumes. Our approach integrates a multi-scale feature fusion network (MFFN) and a synthetic sample-based learning framework (SSLF). The MFFN effectively extracts and fuses multi-scale features under sparse supervision by incorporating channel attention, spatial attention, and a ranking-based loss function. SSLF generates synthetic samples by simply combining fetal head and abdomen data from different fetuses, utilizing semi-supervised learning to improve prediction performance. Experimental results demonstrate that our method achieves superior performance, with a mean absolute error of 166±155 g and a mean absolute percentage error of 5.1±4.6%, outperforming existing methods and approaching the accuracy of a senior physician. Code is available at: https://github.com/Qioy-i/EFW.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1958_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/Qioy-i/EFW

Link to the Dataset(s)

N/A

BibTex

@InProceedings{WanJia_Accurate_MICCAI2025,
        author = { Wang, Jian and Ni, Qiongying and Yu, Hongkui and Yao, Ruixuan and Ying, Jinqiao and Zhang, Bin and Yang, Xingyi and Peng, Jin and Chen, Jiongquan and Yu, Junxuan and Shi, Wenlong and Chen, Chaoyu and Yan, Zhongnuo and Luo, Mingyuan and Cai, Gaocheng and Ni, Dong and Lu, Jing and Yang, Xin},
        title = { { Accurate and Efficient Fetal Birth Weight Estimation from 3D Ultrasound } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15960},
        month = {September},
        page = {35 -- 45}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    This paper proposes a method to estimate fetal birth weight from 3D US images of the fetal head and abdomen shortly before birth. A multi-scale feature fusion network using channel and spatial attention is designed. Synthetic cases are created by joining head and abdomen volumes from different fetusus as unlabeled data to enrich the training set. The method is evaluated on a private dataset of 3D fetal US head and abdomen images acquired 72 hours before delivery and compared to manual FBW computation on 2D US images by measuring head and abdomen circumference by a senior physician.

  • Please list the major strengths of the paper: you should highlight 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 is well written and well structured.
    • The study addresses an important application: The reliable and objective estimation of FBW. The use of 3D US images instead of 2D standard views (which can only be acquired by experienced sonographers) is interesting and worth investigating.
    • Multiple experiments are performed: comparisons to other works, and an extensve ablation study.
    • The synthetic data generation by combining head and abdomen volumes from different patient is simple but apparently quite effective.
  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
    • Using 3D US instead of 2D standard views is interesting, as it is less operator dependent. However, there are limitations of 3D US (lower image quality, lower resolution, not always available). A discussion about these limitations is missing.

    • The authors argue that direct estimation on the 3D volumes instead of computing biometrics is favorable. I disagree. The results are not interpretible any longer. It will be unclear on which grounds a high or low FBW has been estimated. All parents would need some explanation why a cesarian might be necessary. The biometric measurement are very important in this regard. Why not estimating biometrics in 3D? This would indeed provide more information than 2D US. Related to this, the authors hypothesis that “[…] the model predicts FBW by estimating fetal size from both head and abdominal volumes.” However, this is not investigated. This insight into the model and its features is missing. Is the FBW related to head and abdomen size? To offer at least some kind of interpretability, such analysis is important.

    • The comparison to state-of-the-art is not sufficient. The method has only been compared to three other network architectures, which haven’t been designed for the same task. For the two 2D methods, BabyNet and BabyNet++, the authors just report the results from the respective papers on different data. This is clearly not a fair comparison. The study proposes to use different data (3D US instead of 2D US) to estimate FBW. So they have to show that the results are at least comparable to the standart technique (Hadlock in this paper) and automatic methods on 2D data. What they should have done is to record 2D US streams while the physician was manually assessing the fetus and then applying BabyNet on the data. To me, the superiority of 3D US for the task has not been demonstrated.

    • How many years of experience in fetal US has the physician? This information is important. For future work, more physician need to annotate the data. The authors themselves explain that this is an operator dependent task.
    • How many pregnancies and volumes are collected? Does 491 samples refer to pregnancies or 3D US volumes? Are more volumes per pregnancy collected than one head and one abdomen view?
  • 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.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
    • What is the FBW estimation of a synthetic case? A mean of the true FBWs of head and abdomen?
  • 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.

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

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

    The overall writing and structure of the paper is good, and the study proposes an interesting method. However, some important limitations of the approach are not discussed. The comparison to state-of-the-art is insufficient. I’m also not convinced that the presented approach (direct regression of the FBW using 3D US images) is appropriate for the task. Interpretability is important here, and this is not investigated nor discussed.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    I appreciate the clarifications in the rebuttal. If the authors can add some additional discussions on the limitations and use of 3D US and the aim of the method (providing additional information and not replacing 2D US), I can recommend acceptance.



Review #2

  • Please describe the contribution of the paper

    The paper introduces a multi-scale feature fusion network for the estimation of fetal weight. The input of the network is 3D fetal ultrasound screening. The method consists of many existing effective modules, and shows an impressive performance over sota methods.

  • Please list the major strengths of the paper: you should highlight 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 is well-written, and the experiment results are promising.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    The method seems to be a combination of CBAM (ECCV’18), Mamba (COLM’24), and DINO. The authors should clarify the methodological contribution of the paper, or highlight its clinical significance. For instance, is the proposed model real-time? If so, it can enhance the estimation of fetal weight in the clinics efficiently. Is the model larger than the baseline models in parameters?

  • 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 provide sufficient information for reproducibility.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    N/A

  • 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.

    (4) Weak Accept — could be accepted, dependent on rebuttal

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

    Please check the comments above.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Reject

  • [Post rebuttal] Please justify your final decision from above.

    The rebuttal alleviates my concerns on the methodological novelty. However, I am still worried about the comparision of the proposed method with baseline models (e.g., are they with similar parameter size?), as mentioned in my former comments. in addition, multi-scale" and CBAM is not a rare combination, either (e.g.,Multi-Scale TransUnet Combined with CBAM for Nuclear Image Segmentation”). in accordance to all the reasons above, I tend to reject this submission.



Review #3

  • Please describe the contribution of the paper

    The paper introduces a novel deep learning-based approach for estimating fetal birth weight (FBW) directly from 3D ultrasound volumes. This approach, being the first of its kind to leverage 3D ultrasound data for FBW estimation, integrates a multi-scale feature fusion network (MFFN) and a synthetic sample-based learning framework (SSLF). The MFFN is designed to extract and fuse multi-scale features from complex fetal anatomical structures using channel attention, spatial attention, and a ranking-based loss function to enhance prediction accuracy. Meanwhile, the SSLF effectively expands the dataset by generating synthetic samples that improve the prediction performance through semi-supervised learning. Experimental results show that this model achieves superior accuracy compared to existing methods and closely approaches the accuracy of senior physicians.

  • Please list the major strengths of the paper: you should highlight 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: The proposed methodology is novel as it is the first to estimate FBW directly from 3D ultrasound volumes, which provides more anatomical information than the traditional 2D imaging methods. -Innovative Network Design: The multi-scale feature fusion network (MFFN) uses channel and spatial attention mechanisms effectively to integrate features across different anatomical scales and site volumes, which is a fresh approach for feature extraction in this domain. -Synthetic Sample-Based Learning: Introducing the SSLF, which generates synthetic samples by combining fetal data from different sources, is a novel framework that improves the training dataset significantly, addressing the challenge of limited data. -Model Performance: The experimental results demonstrate that the proposed method achieves a mean absolute error of 166.4±154.9 g, outperforming existing methods and nearly matching the accuracy of clinical estimations, showcasing strong evaluation. -Reduction of Operator Dependency: The method effectively minimizes reliance on operator skills by using comprehensive 3D data, which can improve clinical feasibility and applicability. -Comprehensive Ablation Study: The paper provides a thorough ablation study, evaluating the impact of each component of the proposed method, which strengthens the credibility of the results and the proposed methodologies.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    -Limited Comparison with Advanced Techniques: While the paper compares the proposed method with several existing techniques, it does not include comparisons with state-of-the-art architectures for processing spatial-temporal data. Additionally, the comparison to BabyNet++ does not show significant improvements in performance, and the specific advantages of the proposed method over BabyNet++ need to be clarified. -Limited Explainability: The paper lacks discussion on the interpretability of the model. There are no visual explanations (like class activation maps or other visual cues) provided that relate the predicted FBW to specific spatial-temporal features in the 3D ultrasound data. -Real-World Testing: While the method shows promising results in experimental settings, the paper does not provide extensive results from varied real-world clinical environments where ultrasound equipment and protocols may differ. -Clinical Integration: Although the advancements seem promising, the integration into existing clinical workflows might be challenging without further simplification or alignment with standard practices.

  • 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.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    N/A

  • 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.

    (4) Weak Accept — could be accepted, dependent on rebuttal

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

    The paper proposes an original and innovative method for estimating fetal birth weight using 3D ultrasound data, addressing key issues in current clinical practices such as operator dependency and inaccuracy due to spatial limitations in 2D imaging. The integration of multi-scale feature fusion and synthetic sample-based learning frameworks represents significant advancements in the field, demonstrating impressive evaluation results. The comprehensive ablation study further validates the effectiveness of the proposed method. However, the lack of comparison with state-of-the-art techniques, limited real-world testing, the need for additional explainability, and potential challenges in clinical integration are notable weaknesses that need to be addressed. Overall, the paper makes a strong contribution to biomedical imaging, and with further refinement and validation, it has the potential for substantial clinical impact.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    Thanks for addressing my concerns. Overall, this work is novel and interesting. I saw another review also had concerns about interepretability of the proposed method, probably the only weakness that has not been addressed well.



Author Feedback

We thank all the reviewers (R) for reviewing and recognizing our work. Clarifications are provided to address concerns. Q1: Code Release. (R1, R2, R3) A1: Code and 20 test cases will be released upon acceptance for a solid reproducibility, with an interactable website for 3D ultrasound (US) based Fetal Birth Weight (FBW) prediction. Q2: Novelty. (R1) A2: Our work has key innovations in problem definition, methodology, and performance. All reviewers acknowledge that it tackles a basic, long-standing challenge in obstetric US—FBW estimation. We present the first 3D US-based dataset and a precise model, offering a promising alternative to the 40-year-old Hadlock equation. Our method differs from CBAM/MAMBA/DINO, featuring two complementary modules. 1) MFFN boosts feature extraction by integrating multi-scale, -directional channel and spatial attention. Unlike CBAM’s single-scale attention and MAMBA’s one-way modeling, MFFN captures richer spatial cues. The ranking loss further enforces inter-fetal FBW relations. 2) SSFL solves the label shortage. It firstly introduces the pseudo sample generation to enlarge the training set without labels. It further employs a semi-supervised learning to guide knowledge transfer from unlabeled data, unlike DINO’s self-supervised learning. Q3: Clinical Value & Potential. (R2) A3: We thank the Reviewer for the valuable feedback and fully understand the concerns. FBW estimation remains a key unsolved problem. Our clarifications: 1) We never claim that 3D US is superior to biometric methods; our goal is to tackle known 2D manual workflow issues like inter-/intra-observer variation. We thus explore the feasibility of FBW prediction from 3D US volumes, offering a novel view to the field. The proposed 3D auto pipeline is meant as a supplement, not a substitute. We fully acknowledge the clinical importance of biometric parameters for both decision-making and patient communication. Future work will extend our framework to jointly estimate parameters and FBW to improve automation and interpretability. 2) We fully agree with the Reviewer’s concerns about 3D US limitations and will add them in the final version. Still, the strengths of 3D US drove us to explore accurate FBW prediction. The encouraging results support basic feasibility despite known limits. Q4: Comparison. (R2, R3) A4: 1) For SOTA comparison, to our knowledge, no prior work directly estimates FBW from 3D US. So, we chose 3 recent typical 3D regression models trained under similar settings for fair comparison. 2) Direct comparison with 2D methods like BabyNet/BabyNet++ was infeasible due to lack of paired 2D–3D scans in our retrospective data. BabyNet++ needs operator-dependent 2D US videos and expert measurements, introducing inter-observer variability. Future work will add more SOTA 3D models and run prospective studies with paired 2D–3D data for stronger evaluation. Q5: Interpretability. (R2, R3) A5: Visual explanations are omitted due to space limits and will be included in the final version to enhance our hypothesis and interpretability. Q6: Dataset. (R2) A6: Our dataset, collected by two senior doctors with 17 years’ experience, covers 491 pregnancies, each with one head, abdomen and femur scan. Synthetic samples were made by combining head and abdomen volumes from different fetuses. These cases didn’t have ground truths, so we used teacher model predictions as pseudo-labels, which closely match the real distribution. The dataset will be released by the end of 2025. These details will be added in the final version. Q7: Clinical Validation & Integration. (R1, R3) A7: This is the first technical study on direct FBW estimation from 3D US volumes, showing promising feasibility on retrospective data. While full clinical use is beyond scope, we plan future validation and workflow integration. Our model is promising for clinical integration, with real-time efficiency and manageable compute cost (45 ms/case, 100M parameters, 188GFLOP).



Meta-Review

Meta-review #1

  • Your recommendation

    Invite for Rebuttal

  • If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.

    N/A

  • 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



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’

    N/A



Meta-review #3

  • 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



back to top