Abstract

Standard plane localization is a critical prerequisite in ultrasound scanning for disease diagnosis. Although three-dimensional (3D) ultrasound acquires volumetric data superior to two-dimensional (2D) imaging, manually localizing multiple standard planes simultaneously within 3D volume remains labor-intensive and prone to inter-observer variability. To address this, we propose an anatomy-aware multi-task learning framework with structural prior integration and inter-plane geometric constraints to jointly localize multiple standard planes. To validate the effectiveness and generalizability of our method, we curated a multi-center dataset comprising 800 fetal brain volumes. Our approach outperforms state-of-the-art methods, achieving localization errors of 1.09 mm / 3.85°, 1.11 mm / 3.93° and 1.20 mm / 4.03° on three key standard planes of the fetal brain (i.e., the transventricular plane, the transthalamic plane and the transcerebellar plane, respectively). These findings underscore the method’s potential to enhance clinical ultrasound scanning workflows.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2244_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{NieJia_AnatomyAware_MICCAI2026,
        author = { Nie, Jianlong AND Ouyang, Xi AND Zhou, Yuyu AND Wan, Qi AND Xue, Zhong AND Cao, Xiaohuan AND Shen, Dinggang},
        title = { { Anatomy-Aware Standard Plane Localization in 3D Ultrasound with Geometric Constraints } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16893},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper solves automatic standard plane adjustment in 3d fetal US images by using a double encoder structure fusing information from the image and a segmentation mask supplemented by adding geometric constraints for a simultaneous prediction of all 3 standard planes.

  • 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 written comprehensively and the clinical motivation is very clear. Also the dataset and intra-/inter-observer study is a strong point.

  • 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 idea of using the segmentation masks in addition to the image for a downstream task is not entirely new. And it raises the question whether it is really beneficial to have the additional overhead of first predicting a segmentation mask. The paper would have been also stronger if the biometric measurement accuracy would have been reported for all methods and also stating how the requirements are from clinical perspective (to answer whether the proposed method or the other methods are already good enough).

  • 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.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • 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/2026/en/REVIEWER-GUIDELINES.html
    • The baseline seems to be a bit unfair given that the regression does not benefit from the same geom. constraints which will naturally benefit the optimization.
    • it is not clear how errors in the segmentation propagate through the model. This would be easy to test by using the GT vs artificial degradation of the segmentation.
    • It seems to be quite odd how the segmentation template can outperform dedicated prediction models. Do you have a explanation for this?
    • How are the iterative methods initialized?
    • Why was the diffusion model mentioned in the SOTA not used as baseline?
    • Minor: The higher number of digits used to visualize the results in Table 2 is not adding any reasonable context for mm and angle.
    • How time critical is the application as such? Does it need to run on the US scanner with limited resources?
    • I appreciate the ablation study but I would have been interested in seeing how only the geometrical constraints would have performed. without the need of the segmentation mask
    • There are some papers which could be looked at either concerning multi-standard-plane estimation or segmentation as priors: (1) https://arxiv. org/pdf/2305.16717 (2) https://pubmed. ncbi. nlm. nih. gov/33877526/ (3) https://papers. miccai. org/miccai-2025/0071-Paper3576.html
  • 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 — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal

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

    I base my initial recommendation of the limited novelty as the different components used in this study are not entirely and I feel the rationale for doing an additional segmentation is not completely convincing given that it is not clear how the clinical accuracy requirements are.

  • Reviewer confidence

    Very confident (4)

  • [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 still consider the novelty of this paper as rather limited and I am still not entirely convinced of the additional segmentation step. However, I think the evaluation and ablation studies are still solid and I acknowledge the effort the authors put into the rebuttal and will not object to publishing of the paper even though I am borderline leaning towards a rejection. I would like to see that the limitations of an additional segmentation especially when considering degradations in the quality will be mentioned in the camera-ready version.



Review #2

  • Please describe the contribution of the paper

    This paper addresses the automatic localization of 2D standard planes from 3D ultrasound volumes, a task with clear clinical relevance. The method is validated on a sufficiently large dataset, which strengthens the credibility of the results. In addition to evaluating the core task of 2D plane localization itself, the authors further provide indirect validation through downstream biometric measurement tasks. Overall, this is a solid and well-executed piece of work.

  • 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 evaluated on 800 fetal brain volumes, providing strong data support for the experimental evaluation. In the network design, an attention mechanism is used to emphasize masked anatomical regions during image feature extraction, helping the model focus on the most relevant areas within 3D volumes for the localization of 2D standard planes.

  • 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 paper adopts a two-stage framework. In the first stage, a segmentor extracts masks of key anatomical structures; in the second stage, mask information is further injected through both a mask encoder and an attention mechanism. This repeated reliance on mask information naturally raises the question of how the model would perform if the image were no longer used as an input in the second stage. Moreover, a two-stage design inevitably leads to concerns about the overall dependence on the quality of the first-stage segmentation. Perhaps a multi-task end-to-end framework would make the method more compelling and convincing.

  • 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.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • 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/2026/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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal

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

    The paper is evaluated on 800 fetal brain volumes, providing strong support for the experimental results. The proposed two-stage framework is well motivated, using anatomical masks to guide standard plane localization through both a mask encoder and an attention mechanism. However, this strong reliance on mask information raises concerns about the contribution of image input in the second stage and the method’s dependence on first-stage segmentation quality. A multi-task end-to-end framework might make the method more compelling and convincing.

  • Reviewer confidence

    Very confident (4)

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

    N/A

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

    N/A



Review #3

  • Please describe the contribution of the paper

    The manuscript proposes an anatomy-aware multi-task learning framework with structural prior integration and inter-plane geometric constraints to jointly localize multiple standard planes (three fetal brain standard planes) in 3D ultrasound imaging.

  • 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 formulation of multi-plane localization as a multi-task learning problem with explicit inter-plane geometric constraints (parallelism and coplanarity losses) is novel and well-motivated.
    • The introduction of inter-plane geometric constraint losses enforces known anatomical relationships in a novel way and interesting way.
  • 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 segmentation stage is not fully characterized: authors should specify more clearly whether VB-Net is trained from scratch or finetuned, which anatomical structures are segmented and report segmentation performance (such as Dice score). Since plane localization accuracy depends on segmentation quality, reporting these details would strengthen the paper.
    • The evaluation is limited to a single anatomical application (fetal brain), even though the authors discuss this issue and mention it as future work.
    • The clinical motivation for using 3D ultrasound over the more widely available and less expensive 2D ultrasound is not discussed. The practical scenarios where 3D acquisition is justified should be clarified.
  • 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.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    Considering the involvement of human subject data from pregnant women across multiple clinical sites, information regarding ethical approval and patient consent should be explicitly provided.

  • 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/2026/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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal

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

    This paper presents an interesting and well-motivated approach to multi-plane localization in 3D fetal brain ultrasound, introducing novel geometric constraints that encode clinically meaningful inter-plane relationships not exploited in prior work. The experimental evaluation is rigorous, including multi-center data, observer variability analysis, and clinical biometric validation. The primary reason I am not fully confident in giving a strong accept is the absence of explicit information regarding ethical approval and patient consent, which are prerequisites for any study involving human subject data. If the authors can confirm ethical compliance during rebuttal, I would fully support acceptance of this manuscript as a valuable contribution to the conference.

  • 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.

    The authors have addressed my primary concerns, particularly regarding ethics compliance (they confirm that ethical approval was obtained and patient consent was waived by the IRB for this retrospective study). Moreover, the additional details on segmentation and the ablation results demonstrating complementarity between image and mask information strengthen confidence in the methodology, while the clinical discussion for 3D ultrasound (retrospective plane extraction and volumetric measurement for neurodevelopmental assessment) is reasonable and strengthens the paper context. Overall, the study is an interesting and valuable contribution.



Author Feedback

We thank the AC and reviewers for recognizing the significance and strengths of our work, particularly the large dataset (R1, R2), the anatomy-aware attention mechanism (R2), and the multi-task learning framework with inter-plane geometric constraints (R3).

Major concerns:

Q1.Novelty (R1) Although prior studies have utilized segmentation information, our method differs in three key aspects: (1) segmentation masks for both anatomy-aware attention and independent structural encoding; (2) novel inter-plane geometric constraints for simultaneous multi-plane localization; and (3) validation on a multi-center dataset; not explored in the references cited by R1 (we will cite these references). Moreover, our framework is tailored for 3D ultrasound, where severe noise, occlusion, and large search space make localization particularly challenging.

Q2.Segmentation contribution & error propagation (R1, R2, R3) Table 2 demonstrates a clear performance drop when segmentation-related modules are removed. It is further supported by the following ablation results:

Image+geometric constraints: 1.22mm/4.78° (TV), 1.19mm/4.80° (TT), 1.43mm/6.15° (TC) Mask+geometric constraints: 1.17mm/4.36° (TV), 1.21mm/4.51° (TT), 1.33mm/5.12° (TC) Image+Mask+geometric constraints: 1.09mm/3.85° (TV), 1.11mm/3.93° (TT), 1.20mm/4.03° (TC)

Mask information provides stronger localization cues than image alone. The full model achieves the best performance, demonstrating clear complementarity among components..

To mitigate potential error propagation, the localization network is trained using predicted rather than GT masks, improving robustness and ensuring train-test consistency. Robustness analysis using randomly augmented masks (±1mm dilation) shows no significant degradation (p>0.05).

Q3.Two-stage framework design & segmentation details (R2, R3) R2: We adopt a two-stage design because segmentation and plane regression require different feature representations. End-to-end training may lead to suboptimal compromise. It limits search-space reduction, which has been shown to be beneficial in prior work. We will explore this point in future work.

R3: We train a coarse VB-Net from scratch using 9 anatomical labels: skull, CSP, falx cerebri, occipital horn, lateral ventricle (other regions), left cerebellum, cerebellar vermis, right cerebellum, and remaining brain tissue. Dice scores are 0.84/0.71/0.60/0.76/0.64/0.72/0.53/0.70/0.95. Q4.Baseline fairness (R1) Geometric constraints only apply to joint multi-plane prediction and are thus not suitable for single-plane prediction baseline. Multi-plane results (image+geometric constraint) are reported in Q2.Diffusion-based method is excluded in the comparison due to high computational and inference cost. Runtime of our method is 0.25s/case on NVIDIA A4000 and Intel Xeon Silver 4210R.

For the segmentation-based baseline, we first normalize brain position by mask, then estimate planes from the centroids of plane-related labels and inter-plane geometric relations. Its performance benefits from compact standard plane structures and stable geometric relations.

Iterative methods follow the original settings: random plane with warm-up (RL-US); random plane (ITN).

Q5.Clinical application & ethics (R1, R3) R1: In practice, accurate plane localization could lead to more accurate measures, for example, we obtained an average measure error of BPD as 0.43mm, which is much smaller than 1.34mm reproducibility range reported in (Napolitano et al., Ultrasound Obstet Gynecol, 2016). We plan to conduct clinical evaluation for all baselines in future work.

R3: Ethical approval was obtained and patient consent was waived by the IRB for this retrospective study. Compared with 2D ultrasound, 3D ultrasound enables retrospective plane extraction and volumetric measurement, both important for neurodevelopmental assessment requiring cross-plane consistency.

Other minor comments can be properly addressed in the final version.




Meta-Review

Meta-review #1

  • Your recommendation

    Invite for Rebuttal

  • Please justify your decision. 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 is borderline but leans positive: two weak accepts and one weak reject. The reviewers agree that the problem is clinically relevant, the dataset is strong, and the evaluation is more substantial than usual, including observer studies and downstream biometric validation.

    The main concerns are about how much novelty there really is beyond combining segmentation priors with geometric constraints, and whether the extra segmentation stage is fully justified. Several reviewers also ask for a clearer analysis of how much performance depends on segmentation quality, what exactly the segmentation stage contributes, and whether simpler alternatives or fairer baselines would narrow the gap.

    There are also some specific points that should be addressed in rebuttal: details and performance of the segmentation model, fairness of the baseline setup, clarification of initialization and omitted baselines, clinical accuracy requirements for the biometric task, and confirmation of ethics approval and consent.

    The concerns are central enough that rebuttal is appropriate before making a final decision.

  • 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.

    The paper addresses a clinically relevant problem in 3D fetal ultrasound, i.e., automatic localization of multiple standard planes with anatomical and geometric priors. The reviews are overall positive after rebuttal, with two weak accepts initially and Reviewer 3 moving clearly to accept. The work is supported by a comparatively large multi-center dataset, observer variability analysis, downstream biometric validation, and explicit inter-plane geometric constraints.

    The main concerns were the novelty and necessity of the segmentation stage, dependence on segmentation quality, baseline fairness, and missing ethics details. The rebuttal provides useful clarification. In particular, the additional ablations show that image features, mask features, and geometric constraints are complementary, with the full model performing best. The authors also clarify that predicted rather than ground-truth masks are used for training the localization network, report segmentation performance, provide robustness analysis under mask perturbation, and confirm ethics approval with consent waived for the retrospective study.



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.

    The rebuttal addressed the reviewers’ concerns through the necessary clarifications and explanations. Two reviewers supported acceptance, and another revised their assessment to accept after the rebuttal. In my view, the paper addresses a clinically important question, and the evaluation in the original submission is quite solid. Overall, I recommend Accept.



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.

    despite having limited novelty, the reviewers have noted strong clinical motivation. The authors have also addressed the all reviewers’ concerns well with evidence.



back to top