List of Papers Browse by Subject Areas Author List
Abstract
Life-limiting congenital anomalies require accurate prenatal diagnosis for appropriate clinical decision-making.
Prenatal ultrasound (US) examinations involve multiple anatomical planes, and diagnosis depends on identifying anatomical planes and selecting diagnostically relevant planes for each anomaly.
Existing automated methods either rely on plane-level annotations or aggregate heterogeneous images without modeling these diagnostic capabilities.
We propose AnomExpert, a prototype-driven framework for prenatal US anomaly diagnosis using only case-level supervision.
AnomExpert introduces learnable plane prototypes to organize unordered images into latent representations corresponding to anatomical planes without requiring plane annotations.
A disease-aware sparse selection mechanism further selects diagnostically relevant planes for each anomaly.
Experiments on a multi-center dataset of 3,654 cases show that AnomExpert consistently outperforms nine representative multi-instance learning methods.
Using a ViT-small backbone, it achieves 86.9% accuracy and 84.2% F1-score while maintaining parameter efficiency.
These findings indicate that modeling anatomical plane identification and disease-specific plane selection improves weakly supervised multi-plane prenatal US anomaly classification.
The code is available at https://github.com/TIanCat/AnomExpert.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1304_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/TIanCat/AnomExpert
Link to the Dataset(s)
N/A
BibTex
@InProceedings{WanJia_AnomExpert_MICCAI2026,
author = { Wang, Jian AND Yang, Yang AND Pan, Ziheng AND Zhu, Xiliang AND Zhang, Yuhan AND Zhou, Yanfeng AND Ni, Dong},
title = { { AnomExpert: Identifying and Selecting Anatomical Planes for Prenatal Ultrasound Anomaly Diagnosis } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16885},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes a framework for identifying and selecting anatomical planes for prenatal ultrasound anomaly diagnosis using only case-level supervision. A central component of the method is the use of learnable plane prototypes to model informative structure within unordered ultrasound images. The paper’s goal is to improve effectiveness and efficiency relative to existing MIL-based approaches.
- 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.
A key strength of the paper is that the approach practically reduces the need for more detailed annotation.
The application to prenatal ultrasound anomaly diagnosis and the emphasis on anatomical plane selection is clinically relevant.
The paper shows that the proposed method is competitive with other MIL-based baselines while also being efficient.
The inclusion of an ablation study is another positive aspect.
- 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.
One point that would benefit from further clarification is to explain more explicitly in more depth what annotations are available during training and what information is not provided.
A second point concerns the learnable plane prototypes, which are central to the framework but currently somewhat difficult to interpret. It would be helpful to provide more intuition about what a prototype represents in this setting, how the prototype parameters are initialized or learned, and whether these prototypes become associated with particular anatomical planes, anatomical structures, or disease-related patterns after training. The paper would be strengthened by more clearly explaining how this mechanism differs conceptually from clustering latent representations into groups, since at present the distinction is not yet as fully clear as it could be.
Fig. 4 can be somewhat difficult to interpret. In particular, it would help to provide more intuition for why 30 prototypes gives the best performance and why a loss weight of 0.1 yields the highest F1-score. The empirical results are interesting, but additional explanation would them easier to understand.
The separation between contribution 1 and contribution 2 could be presented more clearly. As currently written, it is not entirely obvious whether plane selection is intended as a separate contribution or as an integral part of the overall framework. Clarifying this distinction would help the reader better understand the paper’s contributions.
- 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.
- 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?
I leaned towards a weak accept because the method is effective and efficient relative to MIL baselines on a clinically relevant task, but there is a need for a clearer explanation of the learnable plane prototypes and a more thorough discussion of the empirical results.
- Reviewer confidence
Not confident (1)
- [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 #2
- Please describe the contribution of the paper
Th authors present a novel method called AnomExpert for identifying and selecting anatomical planes from fetal ultrasound for anomaly detection. This method uses learnable prototypes to map latent representations to anatomical planes without the need for annotations. It substantially improves over SOTA multiple instance learning approaches.
- 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 expertly tackles the important issue of learning without expert annotations which ends up being time consuming and requiring expertise. The method is still grounded in clinical knowledge through the learnable prototypes, being weakly supervised at case level. Performing sparse selection of frames that adaptively relate to the disease with substantially better performance than the baselines presents methodologically strong and novel method. An added advantage is that using smaller models does not degrade the performance by much thereby making it useful for resource constrained settings. The results are also clear and well presented in a way that its easy to see the advantage of the method implemented over the baselines.
- 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.
Some aspects of the methodology in section 2 are not clear. For example, in section 2.1 the authors say they introduce K learnable prototypes but do not describe how the prototypes are derived or the initial states before they are learned. The same for the disease queries in eq 5.This affects the reproducibility of the paper.
- Please rate the clarity and organization of this paper
Satisfactory
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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
Some key questions for your method: how do you define the prototypes and disease queries? In the Qualitative results, you indicate that images grouped under the same prototype exhibit consistent anatomical patterns. Is this consistent also with the embeddings?
An experiment that would be an extension of this paper for a journal submission would be to run the method on blind sweep ultrasound videos (data acquired by non-experts). It would be interesting to see if the same performance can be achieved on this data that do not have as clear anatomical views as the standard ultrasound acquired by experts.
From the ablation results in Table 2, dense aggregation without top-k for disease-aware plane selection performs almost as good as the full framework. What is the advantage of adding the top-k selection?
As for the editing, please include better descriptive captions on the figures to summarize the message the figure is supposed to convey
- 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 method presented is novel on technical grounds. However, there are some key aspects on the method description that are not clear- specifically the derivation of the prototype and disease queries.
- Reviewer confidence
Somewhat confident (2)
- [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 paper proposes a supervised framework for anomaly classification that learns anatomical plane prototypes from similar ultrasound views. By explicitly modeling view-specific anatomical consistency and exploiting anatomical view–disease co-occurrence, the method aims to improve discrimination between different disease classes. Experimental results indicate that the proposed approach outperforms state-of-the-art Multi-Instance Learning 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 clearly presented, making the overall contribution easy to follow. +The framework illustration is both clear and visually appealing. +The work demonstrates strong reproducibility: the methodology is thoroughly described, and the authors provide the implementation code (anonymised). +The experimental evaluation includes an extensive and well-structured comparison with state-of-the-art approaches and ablation study.
- 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 terminology “prototype plane” (and the claim of anatomical plane identification) appears somewhat overstated. While the method does learn visually similar views, these do not consistently correspond to anatomical planes or coherent anatomical structures, as suggested by the authors. For instance, Figure 3a shows that Prototype 19 contains both cerebellum and abdominal views and appears to overlap significantly with Prototype 13.This should be addressed by the authors.
+The reported standard deviations in the ablation study are relatively large, with overlapping intervals across different configurations. This weakens the claim that each architectural component/choice yields consistent performance improvements. e.g. the performance difference between Dense aggregation (without top-k) and the full AnomExpert model does not appear to be statistically robust. The authors should clarify how the prototype number (30) was chosen, either through experiments or justified intuition.
- Please rate the clarity and organization of this paper
Satisfactory
- 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.
- 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?
The paper presents an interesting formulation of the problem and interesting framework design.
- 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 authors should address the concerns raised in the weakness section.
- 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?
Overall, the paper offers an elegant formulation of the problem and a thoughtfully designed framework. However, there are several points that require clarification, and addressing these in the rebuttal would strengthen the contribution.
- Reviewer confidence
Confident but not absolutely certain (3)
- [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
Author Feedback
N/A
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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.
Reviewers agreed that this paper presents a clinically relevant and technically strong contribution, offering an effective weakly supervised framework that reduces annotation burden while achieving competitive and efficient performance. The main concerns focus on clearer explanation and interpretation of the learnable prototype and disease queries. Also, the hyperparameter selection needs to be clarified. These clarifications would further strengthen the paper and are recommended for revision.