List of Papers Browse by Subject Areas Author List
Abstract
Standard fetal head plane identification in ultrasound is es-sential for accurate biometric measurements and early detection of neurodevelopmental abnormalities. Automating the identification of these planes can optimize diagnostic workflows. Existing approaches predominantly treat plane identification as a global image-level classification task, often relying on superficial visual features rather than clinically defined anatomical criteria. To address these limitations, we propose SAVer, structure-aware validation framework that couples anatomical reasoning with plane classification to ensure clinically consistent fetal ultrasound plane identification. SAVer integrates anatomical segmentation with plane identification through a unified architecture that combines a perception backbone with a Plane Classification Module along with an Anatomical Reasoning Engine. The reasoning module explicitly evaluates the presence, completeness, and structural consistency of clinically relevant landmarks to ensure alignment with diagnostic standards. In addition, we introduce a Adapter Feature Alignment Module (AFAM) that enables efficient feature alignment between anatomical segmentation representations and plane classification.We evaluate the proposed method on fetal ultrasound head plane data, achieving a structure-size weighted
dice score of 95% for anatomical segmentation and demonstrating improved robustness. The proposed structure-aware framework enhances interpretability and reliability by embedding clinically meaningful constraints into the classification process
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/5513_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/Anushaww26/SAVer
Link to the Dataset(s)
N/A
BibTex
@InProceedings{AmbAnu_SAVer_MICCAI2026,
author = { Amba Prasanna, Anusha AND Ambrose, Vasanth AND Ramesh, Padmini AND Ayyasamy, Shyam AND Ram, Keerthi AND Seshadri, Suresh AND Lakshmanan, Manojkumar AND Sivaprakasam, Mohanasankar},
title = { { SAVer: Structure-Aware Validation of Standard Fetal Head Planes via Clinically Grounded Anatomical Completeness } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16878},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes SAVer, a structure-aware validation framework for fetal standard anatomical plane classification that leverages anatomical information through image segmentation for plane validity discrimination. The method combines a MedSAM-based multi-structure segmentation module with a plane classification branch, alongside an anatomy-aware reasoning to support the validation of standard planes. The main contribution lies in the integration of anatomical structure awareness into the classification pipeline for standard plane detection, which aligns well with clinical practice, where the identification of standard anatomical planes depends on the presence and configuration of specific structures in the image.
- 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 presents the two main strengths. The major strength of the paper relates to the implementation of a complete and new architecture that uses different connected modules for standard anatomical plane detection. Besides having a branch for structure segmentation and another branch for plane classification, the authors implemented an anatomical reasoning module that integrates information from both branches for plane validity estimation, mimicking traditional clinical practice. Despite the segmentation and classification branches not being completely novel, as they integrate state-of-the-art modules and architectures (e. g. , MedSAM, Convolutional Block Attention Module, and block feature adapters), the integration of these modules and the construction of the anatomical reasoning module constitute a well-thought-out architecture. Another strength is the different experiments performed by the authors, validating and evaluating different aspects of the framework (structure segmentation, plane detection accuracy, clinical risk, and ablation studies).
- 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.
Weaknesses can be found in this paper. Firstly, there are missing details in details in the methodology. Although the SAVer architecture is well explained and easy to follow, there are some missing details that are necessary to ensure a complete understanding and reproducibility. In particular, in the Plane-Aware Structured Confidence Aggregation, there are too few details describing the anatomical quality measures (c_struct, c_comp, c_uni, c_dist, c_edge) and how they are calculated. As another example, the authors present the final loss function but do not describe which losses are used for the segmentation and classification tasks. Another weakness is the lack of clarity in the description of the dataset(s) used. The authors begin the methodology section by describing only a dataset with 1,095 samples of the TT, TV, and TC planes. However, in the results section, they mention that the method was evaluated for Standard Plane Validity Assessment on a testing set of 240 images, including both standard and non-standard planes. Subsequently, the Clinical Reliability Analysis is evaluated on a validation set, and Figure 4 seems to suggest that this set contains only 60 images (with only 5 corresponding to non-standard planes). Finally, it is unclear on how many images the anatomical segmentation performance was evaluated (e. g. , whether it was performed on the 1,095 samples using cross-validation). Finally, another weakness is the lack of comparison to the state of the art, as the authors do not compare their method with other available approaches.
- 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 does not provide sufficient information for 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
- There are some sentences that seem to be more related to assumptions than to evidence. For example, the sentence on page 2, “Without a structure-aware framework, the model is prone to make high-confidence misclassifications, often predicting general elliptical head contours for a standard plane. ” This type of statement should be supported by evidence (i. e. , references) indicating that such methods focus mostly on head contours. This may constitute an overstatement.
- Another example of such sentences: On page 6, the authors argue, “While the False Positive Rate (FPR) is slightly higher (0.20–0.36), this reflects a clinically desirable conservative bias that favors accepting potentially valid planes over discarding diagnostically useful frames. ” This can be true in a diagnostic context, where it may be preferable to have false positives rather than false negatives. Nevertheless, in the case of plane detection, since biometry is performed on these planes, it can be argued that it is preferable for the method to accept only planes that are certainly valid, rather than frames that may be valid. These statements could be expressed in a more cautious manner.
- More details about how the anatomical quality measures are calculated will strengthen the manuscript. For example, how is the quality for intra-structure distribution measured? And how is uniformity defined? For improved clarity and reproducibility, the authors should clarify this.
- As stated in the weaknesses, it is not clear how the training, testing, and validation datasets are defined and if they are constant for all the experiments. This is an important aspect of the work and should be clear.
- In Table 2, the authors present IoU with units of (mm) and a downward arrow. IoU should be a dimensionless value ranging from 0 to 1 (or 0–100%). Moreover, the objective is to achieve a high IoU. Please rectify this.
- 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?
The paper presents an interesting and clinically relevant pipeline. The overall design of the framework combines multiple components in a meaningful way. However, the weaknesses identified prevent a straightforward acceptance. That said, some of these concerns, especially those related to clarity and missing details, could potentially be addressed in the rebuttal. For these reasons, while the paper currently falls slightly below the acceptance threshold, it could be considered for acceptance if these issues are adequately clarified or addressed during the rebuttal process.
- Reviewer confidence
Somewhat confident (2)
- [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 responded convincingly to the limitations and concerns raised in the initial review. In particular, they provided additional methodological details regarding the points previously highlighted and presented a clearer and more structured description of the datasets and their use across the different experiments, namely clarifying the training, validation, and testing splits.
The authors argue that these clarifications improve the quality and transparency of the manuscript, and so the reviewer assumes that these modifications will be properly incorporated into the final version of the paper. In addition, the authors also addressed minor concerns, such as statements that appeared to be based more on assumptions than evidence, indicating that these sections will be revised accordingly, as well as correcting the typos identified in the tables.
Overall, the reviewer feels comfortable recommending the acceptance of this paper.
Review #2
- Please describe the contribution of the paper
The paper proposes SAVer, a unified framework that goes beyond simple image-level classification by combining plane classification with anatomical segmentation and reasoning to ensure that predicted standard planes are clinically valid. SAVer introduces an Anatomical Reasoning Engine that checks the presence, visibility, completeness, and spatial relationships of required landmarks without retraining the full encoder. The results also show that SAVer can segment 32 clinically relevant inner fetal brain structures accurately.
- 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.
1.A major strength is that the paper does not treat fetal standard-plane recognition as only a generic image classification problem. Instead, it reformulates the task around clinical anatomical criteria, which is much more aligned with how ultrasonographers assess validity. 2.SAVer has novel integration of segmentation, classification, and reasoning, which is interesting. 3.Another strong aspect is the Anatomical Reasoning Engine, which explicitly checks whether relevant anatomical landmarks are present and structurally consistent.
- 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.
1.This work considers 32 brain structures with varying sizes. How do these fine-grained structures contribute to the SAVer framework? Are all of them necessary? Would using only larger or more significant structures achieve similar performance? Also, annotating 32 structures is very costly. 2.A dependency matrix M_{c,k} is defined, but it is unclear how it is constructed. Is it based on clinical knowledge or defined by clinicians? This matrix should be better explained. 3.Fig. 2 shows the Adapter module, but components such as the DRL module, Uncertainty Head, and Blend are not described in the Methodology section. 4.Will this dataset be publicly released? Also, how are the training, validation, and test sets split? 5.Are the weights w_i fixed and equally distributed, or are they dynamically assigned? 6.In the ablation study, which baseline is used? Is it MedSAM or another transformer/CNN model? 7.Fig. 2 reports a value (0.6975) for partially standard planes. Does SAVer classify SP, non-SP, and partially SP? If it only distinguishes SP and non-SP, this figure is confusing. 8.There are some minor issues: “SoftMax” should be “Softmax”; an indent is missing after Equation 12; and several sentences have missing spaces between words.
- 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 does not provide sufficient information for 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.
(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?
The paper presents an interesting idea by integrating segmentation, classification, and reasoning into a unified framework, which is novel and potentially valuable. However, the main concern is the clarity of the paper. The writing needs significant improvement to make the method easier to understand and follow. Several components are not well explained, and important details are missing, which makes it difficult to fully assess the approach. Overall, while the idea is promising, the current writing limits its impact. I would be happy to raise my score if these concerns are addressed in rebuttal.
- 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.
After reading the authors’ rebuttal, I would like to change my rating to Accept. The rebuttal addressed most of my concerns satisfactorily. I still encourage the authors to further improve the writing and clarity of the paper in the camera-ready version. It would also be valuable to include more discussion on the clinical translation of the proposed method and outline possible future work directions.
Review #3
- Please describe the contribution of the paper
This paper focuses on three fundamental cranial standard planes in 2D ultrasound and uses anatomical structure segmentation to assist in distinguishing standard from non-standard planes. Its contribution lies not only in the design of a classification method, but also in treating the confidence estimation results as an important basis for guiding the clinician’s next actions, together with a clinical risk and uncertainty analysis to evaluate the reliability and safety of the method in clinical practice.
- 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.
This paper builds an end-to-end joint segmentation and classification framework based on a fine-tuned MedSAM, and further incorporates medical priors regarding the anatomical structures that should be present in different standard planes to design task-specific constraint losses. By introducing explicit anatomical constraints into the learning process, the method not only improves interpretability and task adaptability, but also shows good extensibility, making it easier to generalize to other plane recognition tasks and related medical image analysis problems. Given that ultrasound imaging is highly dependent on operator experience, accurate discrimination between standard and non-standard planes remains of great clinical importance.
- 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 experimental section lacks sufficient comparisons with other classification models. Considering that there are already multi-task methods in the classification literature that integrate segmentation and classification, the current comparative experiments do not provide sufficiently strong evidence to convincingly demonstrate the advantages of the proposed method.
- 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?
This paper proposes an end-to-end joint segmentation and classification framework based on fine-tuned MedSAM, incorporating anatomical priors from different standard planes to design task-specific constraint losses. The method improves interpretability and has potential for extension to other plane recognition tasks, which is clinically meaningful given the strong operator dependence of ultrasound imaging. However, the experimental evaluation is not sufficiently convincing, as it lacks adequate comparisons with other classification or segmentation-classification multi-task models.
- 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
Author Feedback
We thank R1, R2, R3 for their constructive reviews. R2 confirmed the method is novel and clinically translatable; R1/R2/R3 acknowledged strong performance on challenging cases. Comparison with Existing Architectures(R1,R2,R3):Our core claim is that classification-only models are insufficient for standard plane validity assessment. The Only classification(CLS) baseline (Tab.3) uses the identical ViT-B encoder as MedSAM — a strong transformer pretrained on large-scale medical data yet achieves F1 71.70% with FNR 0.31, incorrectly rejecting 31% of valid standard planes. SAVer achieves F1 93.95% and FNR 0.03: a +22.25 F1 gain and 10× FNR reduction. Critically, structural completeness supervision (CLS+Seg+Struct) alone drops FNR from 0.41 to 0.06 (85% reduction), confirming classification fails where global appearance is misleading. While prior MTL methods (Lin et al.,2019; Zhang et al., 2021) address fetal US quality via structure detection and binary classification, no model anatomical completeness, spatial consistency, or confidence-graded scoring across near-standard planes making the ViT-B ablation the most rigorous comparison of our core contribution. Aggregation Metrics and Loss Functions(R1,R2): All five metrics are clinically motivated and orthogonal.c_struct: mean pixel-wise segmentation confidence within detected regions. c_comp: presence and sufficient area of plane-specific structures with hard constraints on mandatory anatomy (e.g., CSP for TT plane). c_uni: coefficient of variation of structure areas, measuring anatomical balance. c_dist: quadrant-wise spatial distribution, penalizing implausible configurations. c_edge: penalizes border-intersecting structures, enforcing centered framing per clinical standards. Together these cover completeness, certainty, balance, spatial plausibility, and framing aspects no classifier can model. L_seg = Dice + Focal + Soft HD + L2; L_cls = Cross Categorical Cross-Entropy. The DRL module, Uncertainty Head, and Blend operation form the Adapter for lightweight task-specific adaptation; full equations will be added in the revision. Design Choices(R2): The 32 clinician-recommended structures provide complete discriminative coverage across TC, TT, TV planes. Near-standard planes differ not in global appearance but in subtle structural absences or spatial misplacements invisible to classifiers validated by the ablation: structural supervision improves F1 from 81.96% to 88.24% and reduces FNR from 0.41 to 0.06 over CLS+Seg. Annotation is a one-time dataset cost with no impact on inference. M_{c,k} is clinician-defined, encoding mandatory vs. optional structures per plane (e.g., CSP and thalami mandatory for TT), directly reflecting acquisition protocols. Weights w_j are fixed hyperparameters from iterative validation tuning; values will be reported in the revision. Confidence scores: Standard (>0.8), Partially Standard (0.55-0.8), Non-Standard (<0.55); both latter categories are clinically invalid, ensuring no borderline plane reaches downstream tasks. Dataset(R1,R2): Training: 1,095 images (TC, TT, TV). Validation/Test: 240 images (89 TC, 92 TT, 60 TV).Structure validation was also performed by clinicians across all three planes.Clinical Reliability: 60 TV images (55 standard, 5 non-standard). Splits are consistent across all experiments and will be clarified.Code will be publicly released. Writing and Claims(R1): Overstatements will be revised to evidence-grounded language citing ablation results. The FPR claim on page 6 will be reframed to acknowledge that specificity is equally critical in biometry-dependent plane selection. The IoU error in Tab.2 is corrected — IoU is dimensionless and higher is better. The 10x FNR reduction and +22.25 F1 gain over the strongest available classification backbone demonstrate a fundamentally more capable framework. We are confident the outlined revisions will substantially strengthen the manuscript.
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.
The main issue is clarity and completeness of the method description. Key components are not sufficiently specified, including the anatomical quality measures, the dependency matrix, and several modules shown in the figures but not described in the text. This makes it difficult to assess the approach and raises concerns about reproducibility.
Dataset: The splits between training, validation, and test sets are not clearly defined, and there is inconsistency in the reported dataset sizes across experiments. In addition, the lack of comparison to relevant state-of-the-art methods, particularly segmentation-classification or multi-task baselines, weakens the empirical claims.
There are also questions about the necessity and contribution of some design choices, such as the use of a large number of anatomical structures and how these are weighted and integrated into the reasoning module.
A rebuttal should focus on making the methodology fully explicit, clarifying dataset usage, and strengthening experimental comparisons. The final decision could go either way depending on how convincingly these points are addressed.
- 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 addresses the main concerns on method clarity, dataset splits, and the role of the anatomical reasoning components. Both weak-reject reviewers moved to Accept, while the remaining reviewer was already Weak Accept. The paper is clinically motivated, reasonably novel in its structure-aware formulation, and now sufficiently clarified.
Camera-ready should still improve writing, explicitly define all aggregation metrics and weights, clarify the dataset splits, and include the corrected IoU notation.
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.
All three reviewers recognized the clinical relevance and novelty of integrating segmentation, classification, and anatomical reasoning into a unified standard-plane validation framework. The rebuttal addressed reviewers’ concerns and two reviewers moved to accept post-rebuttal. The recommendation is therefore 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.
Reviewers are aligned positively after rebuttal and find the paper clinically meaningful with a satisfactory methodological contribution. The rebuttal clarified the dataset split, methodological details, and limitations, and the updated reviewers indicate that their main concerns were addressed.