List of Papers Browse by Subject Areas Author List
Abstract
Accurate segmentation of the fetal head (FH) and pubic symphysis (PS) in intrapartum ultrasound is crucial for Angle of Progression (AoP) estimation. However, speckle noise, weak boundary contrast, and anatomical variability hinder segmentation. We propose DMC-Net, a dual-domain meta-conditioning framework that integrates structural and spectral priors in a stage-specific manner. The encoder uses a Structural Domain Meta Conditioning Module to inject anatomical cues, reducing noise and enhancing boundary coherence. A Frequency-aware Cross-scale Attention Module in deeper layers models spatial-frequency interactions, suppressing noise while maintaining structural continuity. In the decoder, a second DMCM stabilizes reconstruction with refined frequency information. Finally, a Progressive Spectral Reconstruction Decoder restores Fourier domain components for precise boundary recovery. Experiments on the PSFHS2023 dataset show DMC-Net outperforms state-of-the-art methods in segmentation accuracy and AoP estimation, demonstrating its effectiveness in challenging ultrasound scenarios. Our code and data are available at https://github.com/XingLongH/DMC_Net.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/5136_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{HuXin_DualDomain_MICCAI2026,
author = { Hu, Xinglong AND Dong, Lan AND Zhang, Lu AND Ye, Zhaoyi AND Wang, Du AND Mei, Liye AND Lei, Cheng},
title = { { Dual-Domain Meta Conditioning Network for Fetal Head and Pubic Symphysis Segmentation in Ultrasound Images Analysis } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16883},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
Propose DMC-Net, a dual-domain meta-conditioning network that explicitly integrates structural and spectral priors for fetal head (FH) and pubic symphysis (PS) segmentation in intrapartum ultrasound, addressing speckle noise, weak boundaries, and anatomical variability. Design the Domain Meta Conditioning Module (DMCM) that injects anatomical or frequency meta-information to guide query modulation, enhancing structural consistency and reducing noise amplification. Introduce the Frequency-aware Cross-scale Attention Module (FCAM) that combines multi-scale modeling and frequency decomposition to suppress high-frequency noise while preserving critical edge details. Develop the Progressive Spectral Reconstruction Decoder (PSRD) that refines magnitude and phase in the Fourier domain for precise boundary recovery and accurate Angle of Progression (AoP) estimation. Validate superior performance over state-of-the-art methods (CMU, Mamba-Sea, U-KAN, etc. ) on the public PSFHS2023 dataset, with improved segmentation accuracy and reduced AoP error.
- 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.
High clinical relevance: The work targets a critical clinical task for labor progression assessment and delivery mode decision, solving the subjectivity and high variability of conventional vaginal examinations. Strong methodological novelty: DMC-Net breaks through single spatial-domain modeling and uses a structural–frequency dual-domain paradigm well-matched to ultrasound imaging characteristics. The DMCM, FCAM, and PSRD are well-motivated and technically distinct from standard U-Net or Transformer variants. Rigorous and complete evaluation: Experiments are conducted on the standard PSFHS2023 challenge dataset with comprehensive comparisons and ablation studies. Quantitative metrics (DSC, HD, ASD, ΔAoP) and visual results consistently support the effectiveness. Clear and reproducible pipeline: The network architecture, module formulations, training settings (optimizer, lr, batch size, GPU), and implementation details are well-described. Robust performance in challenging conditions: The model maintains stable segmentation under severe speckle noise, acoustic shadowing, and fetal motion, showing strong potential for clinical translation.
- 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.
Lack of efficiency analysis: The paper does not report model parameters, FLOPs, or inference speed, which are essential for real-time clinical ultrasound deployment. Insufficient generalization validation: Experiments are limited to the PSFHS2023 dataset; no cross-device, cross-center, or external dataset validation is provided. Partial novelty clarification: While the integrated pipeline is novel, frequency-domain attention and spectral reconstruction have been explored in prior medical image segmentation works. The paper could better highlight its unique design choices. Limited clinical case analysis: The paper does not analyze typical failure cases (e. g. , severe head rotation, strong acoustic shadowing) from a clinical perspective. Minor presentation issues: Some mathematical symbols lack clear definitions, and ablation studies lack corresponding visual comparisons.
- 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.
(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?
This paper addresses a high-impact clinical task with a novel, well-designed dual-domain meta-conditioning framework. The method is technically sound, experiments are comprehensive, and results are superior to recent state-of-the-art approaches. Although minor improvements in efficiency and generalization validation are needed, the core contributions are solid and significant for intrapartum ultrasound analysis. The paper fully meets MICCAI acceptance criteria.
- 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.
After reading the authors’ rebuttal, I recommend acceptance. My initial concerns were mainly about efficiency analysis, generalization validation, novelty clarification, and clinical failure-case discussion. The rebuttal provides reasonable clarifications and sufficiently addresses these issues at the current conference-paper stage. The core contribution remains technically solid: the paper proposes a clinically meaningful dual-domain meta-conditioning framework for fetal head and pubic symphysis segmentation, with well-motivated structural and frequency-domain modeling. The experimental results on PSFHS2023 are competitive, and the ablation studies support the effectiveness of the proposed DMCM, FCAM, and PSRD components. Although external validation and further lightweight optimization would strengthen the work, these limitations do not undermine the main contribution. Overall, I believe the paper meets the acceptance standard.
Review #2
- Please describe the contribution of the paper
This paper proposes DMC-Net, a dual-domain meta-conditioning network for fetal head and pubic symphysis segmentation in intrapartum ultrasound, aiming to achieve reliable Angle of Progression estimation. The main contributions are: (1) The authors design DMCM, FCAM, and PSRD to enhance structure awareness, suppress noise, and recover fine boundaries. (2) Experiments on the PSFHS2023 dataset demonstrate that the proposed method achieves competitive performance.
- 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) This study focuses on the segmentation of the fetal head and pubic symphysis in intrapartum ultrasound, which has clear practical clinical significance. (2) The method targets typical challenges in ultrasound images, including speckle noise, weak boundaries, and anatomical variations, with a complete and logical overall framework. (3) The proposed method achieves competitive results on the public PSFHS2023 dataset, with improvements in both segmentation accuracy and AoP estimation.
- 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) The proposed method represents an incremental improvement upon the U-Net architecture. However, the added components: including the meta-information-based cross-attention-style design (DMCM), multi-scale fusion-style mechanism (FCAM), and frequency-domain processing-style module (PSRD), are all widely adopted in existing literature [1,2]. As such, the work essentially constitutes a straightforward combination of established architectural designs, rather than introducing fundamentally novel mechanisms. (2) It remains unclear what specific input constitutes the “meta-info” in the DMCM. Is it a random learnable parameter, or a structured anatomical prior derived from the data? This critical design detail is insufficiently clarified. (3) While the ultimate goal of the work is AoP estimation, the model itself is fundamentally a segmentation network. The lack of validation on multiple independent datasets makes it difficult to substantiate the claimed competitive improvements in clinical utility. Furthermore, given that the authors have stacked multiple complex modules, it is also unclear whether such a sophisticated model is suitable for real-world clinical deployment.
[1] Zhou Z, He A, Wu Y, et al. Spatial-frequency dual domain attention network for medical image segmentation[C]//2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). IEEE, 2024: 4076-4081.[2] Zhou Y, Huang J, Wang C, et al. Xnet: Wavelet-based low and high frequency fusion networks for fully-and semi-supervised semantic segmentation of biomedical images[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2023: 21085-21096.
- 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 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 aims to perform segmentation of the fetal head and pubic symphysis in intrapartum ultrasound, which has clear practical clinical significance. The overall structure and logic of the paper are well organized. However, the proposed model is merely an incremental improvement formed by stacking multiple existing modules, and only a small number of datasets are used for validation in the segmentation task.
- 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 made efforts to address the concerns. While some minor issues remain, I maintain my original decision.
Review #3
- Please describe the contribution of the paper
The paper proposes DMC-Net, a segmentation framework for intrapartum ultrasound that improves fetal head and pubic symphysis delineation by combining structural priors and frequency-domain priors through dual-stage meta-conditioning.
- 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 main novelty is a dual-domain meta-conditioning framework that explicitly combines structural/anatomical cues with frequency-domain priors at different stages of the network, rather than relying only on standard spatial CNN features. This is interesting because ultrasound quality is strongly affected by both structural ambiguity and speckle/high-frequency artifacts, so modeling both domains is well motivated.
- 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.The choice of scale factors s \in {4, 2} in the FCAM module is not well justified. Please explain why these values were selected and provide an ablation study to demonstrate their effectiveness for PS and FH segmentation. 2.There is an inconsistency in the dataset description. The PSFHS2023 dataset [4] (Zenodo) reports 1,358 images from 1,124 pregnant women, while this study uses 4,000 training, 400 validation, and 700 test images. This discrepancy should be clearly explained, and the correct reference should be verified. 3.The details of the initial convolution block \phi_0(I) are unclear to show how it extracts meta-information. What are the details of the convolution layers? 4.Important experimental details are missing, such as the number of training epochs, data augmentation strategies, and model complexity (e.g., number of parameters and FLOPs). These should be included for reproducibility. 5.The proposed architecture integrates multiple specialized modules, which increases model complexity. This may affect robustness, computational cost, and memory usage, especially in real-time clinical settings such as labor wards. 6.How about the runtime performance or parameter efficiency? There is no such clear info to claim the practical deployability and real-time applicability. 7.Although the paper focuses on segmentation, the clinical motivation is AoP measurement and labor outcome assessment. In Table 1, the improvement in AoP (0.02) appears marginal. In Fig. 2, the predicted masks do not show clear improvement over U-KAN. The clinical value of the proposed method remains unclear and should be better justified.
- 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.
- 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 main concerns are the limited clinical impact and weak reproducibility. Although the method is technically complex, the improvement in AoP estimation is marginal and does not clearly translate into meaningful clinical benefit. In addition, some experimental details are missing, making the work difficult to reproduce. The paper also does not report runtime or efficiency, which is important since AoP estimation is expected to be used in real-time clinical settings. As a result, the practical applicability of the method remains unclear.
- 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 would like to change my rating to Accept, as the authors addressed most of my concerns satisfactorily. However, I would still encourage the authors to include the following in the camera-ready version: report mean/std and p-values for AoP measurements to better demonstrate the statistical significance of the improvement; include computational efficiency metrics, such as parameter count and inference speed; provide a clearer and more detailed description of the “meta-info” component.
Author Feedback
We sincerely thank all reviewers (Meta, R1–R3) for their constructive and insightful feedback. We are encouraged by the reviewers’ recognition of the clinical significance (Meta, R1, R2), novelty (R1, R3), well-motivated methodology (R1, R3), and technical contributions (R1, R2) of our work. Below, we address the main concerns. 1.Novelty and “Meta-Info” Clarification (Meta, R1, R2, R3): Our framework introduces a task-oriented dual-domain collaborative design. The core novelty lies in a progressive dual-domain meta-conditioning paradigm, where complementary structural and frequency priors are hierarchically injected at different optimization stages for intrapartum ultrasound segmentation, rather than relying solely on feature fusion or self-attention. Specifically, “meta-info” in DMCM refers to task-specific prior features extracted from different domains. In the encoder stage, shallow structural features are injected into the decoding process through the first DMCM to improve anatomical consistency. In the decoder stage, FCAM models spatial-frequency interactions to generate frequency-aware priors, which are injected into a second DMCM to refine reconstruction and recover fine-grained boundaries. PSRD further refines amplitude and phase components using enhanced frequency-aware priors, recovering high-frequency details while suppressing low-frequency artifacts. The ablation experiments in Table 2 also demonstrate the performance improvement brought about by the utilization of prior information and module design. We will further clarify the design motivation and distinctions from existing methods. 2.Experimental Details and Reproducibility (Meta, R3): To ensure reproducibility, we will release details such as the source code, training configuration, and hyperparameter settings upon publication. 3.Dataset Description (Meta, R3): The dataset follows the PSFHS Challenge introduced in MICCAI 2023 [4], which consists of the PSFHS and JNU-IFM datasets. The Zenodo page reports statistics only for the PSFHS datasets, whereas our study uses the complete challenge dataset containing 5,100 ultrasound images. 4.AoP Measurement (Meta, R3): Although AoP improvement over U-KAN appears small, AoP is highly sensitive to boundary localization accuracy of both fetal head and pubic symphysis. Even minor segmentation improvements can enhance measurement stability and reduce clinically relevant localization errors. Moreover, our method achieves better DSC, HD, ASD, and Score metrics than U-KAN, indicating more reliable segmentation performance. 5.Computational Efficiency (R1, R3): The proposed model contains 44.405M parameters and requires 95.692 GFLOPs. It achieves real-time performance at 52.78 FPS on an NVIDIA RTX 3090 GPU, with an average inference time of 18.946 ms per image. 6.Insufficient Generalization Validation (R1, R2): Our data come from two independent datasets involving multiple hospitals and clinical environments. This introduces variability in imaging quality, patient distribution, and acquisition conditions, providing a certain degree of generalization evaluation. 7.Failure Cases (R1): Failure cases mainly occur when fetal head boundaries are severely blurred or affected by strong acoustic shadowing. These cases highlight the importance of structural and frequency priors for robust ultrasound segmentation. We will present the failed cases in the qualitative results. 8.FCAM Scale Factor (R3): The scale factors balance global context and local detail. Larger downsampling captures global anatomical structure, while smaller scales preserve fine-grained boundary details. 9.Details of \phi_0(I) (R3): The initial block \phi_0(I) uses lightweight convolutional layers to extract low-level structural features from ultrasound images. Such features better preserve anatomical topology and edge information. We will further provide architectural details including kernel sizes and channel dimensions in the revised 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 reviewers agreed the paper solves a clinically meaningful problem and the framework design is interesting. However, concerns remain regarding the degree of methodological novelty, reproducibility, and experimental details.
The rebuttal should address the following points, including but not limited to: 1.Methodological novelty. Clarify how the proposed modules differ from prior methods of frequency-domain and dual-domain segmentation.
2.Experimental details and reproducibility. As the approach is complicated, will the code be shared? Also, provide the missing details such as dataset details (the discrepancy in reported samples), training details, hyper-parameter choices, etc.
3.Justification of AoP estimation improvement. As the AoP measurement is the clinical motivation, the improvement is marginal. The discussion should be provided.
- 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 satisfactorily addressed the major concerns raised during the rebuttal. Remaining concerns are mainly minor and relate to strengthening the final presentation, such as providing stronger statistical evidence (e.g., mean/std, p-values), reporting computational efficiency metrics, and improving descriptions of some methodological components. These issues are useful suggestions for the camera-ready version but do not undermine the core contribution of the paper. Overall, the work presents a technically sound and clinically relevant contribution with convincing experimental support. Therefore, I support Accept.
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 active reviewers moved to accept post-rebuttal and therefore, 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.
The proposed dual-domain meta-conditioning framework is recognized as technically sound and highly clinically relevant for intrapartum ultrasound analysis. While initial concerns were raised regarding architectural novelty, meta-info definition, and efficiency metrics, consensus among reviewers post-rebuttal acknowledges that the authors have successfully addressed these issues.