List of Papers Browse by Subject Areas Author List
Abstract
Accurate localization of acupoints in ultrasound imaging is critical for safe and effective acupuncture, yet remains challenging due to low tissue contrast, speckle noise, and significant inter-subject anatomical variability. Existing methods often rely on implicit texture learning or rigid landmark detection, failing to account for the depth-dependent physiological nature of the Deqi response.
To address these limitations, we propose the Physio-Anatomical Probability Field Network (PAPF-Net), a novel framework that shifts from deterministic point regression to uncertainty-aware distribution modeling.
The core technical contribution lies in synergizing two complementary mechanisms. To address anatomical heterogeneity, the Category-Conditional Anatomical Prototype (CCAP) module explicitly incorporates site-specific priors by deforming learnable templates to conform to individual anatomical geometries. Concurrently, to mitigate acoustic ambiguity, the Uncertainty-Aware Decoder transitions from deterministic regression to probabilistic modeling, capturing signal attenuation in deep tissues through distribution-aware Sinkhorn supervision.
Validated across a large multi-center cohort, our approach surpasses current standards in both accuracy and robustness. Its focus on neuro-fascial interfaces ensures physiological plausibility, offering an interpretable, standardized basis for ultrasound-guided acupuncture.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1526_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{DaiGuo_PhysioAnatomical_MICCAI2026,
author = { Dai, Guowei AND Dai, Duwei AND Zhang, Yi AND Ji, Yulong AND Chen, Hu},
title = { { Physio-Anatomical Prior-Guided Probabilistic Field Learning for Ultrasound-Based Neuro-Fascial Interface Landmark Localization } },
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
The main contribution of this paper is the proposed PAPF-Net, which reformulates ultrasound acupoint localization as physio-anatomical probabilistic field learning. By integrating deformable anatomical priors with uncertainty-aware modeling, it enables more accurate and physiologically plausible localization of neuro-fascial interfaces in challenging ultrasound images.
- 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 strong in its clinically motivated formulation, reframing ultrasound acupoint localization as probabilistic field learning rather than point regression. It also introduces a meaningful use of deformable anatomical priors to handle anatomical variability, and is supported by a solid evaluation on a multi-center dataset.
- 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 main weakness is the limited methodological novelty, as the framework mainly combines existing ideas such as anatomical priors, deformable alignment, and uncertainty-aware prediction. In addition, the evaluation is restricted to three acupoint categories, leaving its broader generalizability insufficiently validated.
- 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.
(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?
My recommendation is based on the paper’s clinical relevance and solid empirical performance, balanced against its limited methodological novelty. While the proposed framework is well motivated and effective for the target task, its technical contribution appears largely incremental, and the evaluation scope is still somewhat limited in terms of broader generalizability.
- 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 rebuttal successfully addresses the main concerns regarding reproducibility and methodological clarity by explicitly defining the optimization objective, prototype mechanism, auxiliary supervision, and the aleatoric nature of the uncertainty modeling. While the individual components are not entirely novel in isolation, the paper’s contribution lies in the clinically motivated integration of site-conditioned anatomical prototypes, deformable alignment, and probabilistic supervision for ultrasound-guided Deqi response localization. The experimental evaluation is sufficiently comprehensive for the stated scope, including multi-center data, multiple anatomical sites, strong baselines, and detailed ablations, which together support the effectiveness of the proposed framework. Although the methodological novelty is moderate and broader anatomical generalization remains future work, the remaining issues are primarily related to presentation rather than technical validity. Overall, the rebuttal resolves the major concerns and supports acceptance of the paper.
Review #2
- Please describe the contribution of the paper
The author(s) propose an uncertainty-aware learning framework based on Probabilistic Field Learning that reflects location-specific reliability to resolve existing deterministic point regression method’s vulnerability to ultrasound signal attenuation, noise, and anatomical variability between subjects. They attempt to solve the problem of anatomical heterogeneity by introducing a categorical conditional anatomical prototype module that adapts to individual geometric structures instead of a fixed template, and ensure alignment accuracy between the predicted distribution and the actual physical structure by applying a distribution-aware loss function—Sinkhorn—to target a physiological indicator called the neuro-fascial interface.
- 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.
Major strengths: 1.End-to-end guided learning through site prototypes and warping appears to be a novel methodology suitable for learning that reflects physiological diversity. This research may contribute to the expansion into interpretability and precision medical diagnosis. 2.Significantly improved performance resulting from proposed model architecture.
- 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.
Major weaknesses: 1.For reproducibility, the forms of NLL, EMD (Sinkhorn), AUX Loss, and Label(or site indicator) must be clearly described/illustrated. 2.It is unclear how learning regarding uncertainty was performed. It appears that a surrogate Gaussian distribution was used, but this should be clearly specified. 3.The uncertainty asserted in this study is closer in nature to “epistemic (model-derived)” rather than “aleatoric (data-intrinsic).” No grounds or methodological arguments supporting the claim of it being “aleatoric” have been revealed. Given that this concerns the nature of the study’s main keyword—uncertainty—it must be clearly clarified. 4.Insufficient reproducibility.(The main contribution of this study should be considered its strong performance, but verification of this is difficult.) 5.Despite the research value of the methodology, simple metric-based analyses were performed on a single task. To be valuable as research, in-depth analysis of the subject or discussion of various tasks is required. 6.Isn’t ALHR a precision score for segmentation labels? Without comparative verification with other indicators or a discussion regarding the specificity of expert annotations, it seems difficult to consider the metric as having been “propose”d.
- Please rate the clarity and organization of this paper
Poor
- 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 author(s) propose a novel methodology that reflects physiological diversity. They support its rationality by demonstrating the resulting performance improvements and ablation studies. Nevertheless, experimental performance enhancement must be predicated on reproducibility. If experimental reproducibility is lacking, the rationality and structural necessity of the methodology must at least be demonstrated. Without data (codes or supplementary materials) to support reproducibility or sufficient arguments regarding its necessity, it is difficult for the proposed method to be considered valid research, even if it is novel. (Ref. Major weaknesses)
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Reject
- [Post rebuttal] Please justify your final decision from above.
Concerns regarding uncertainty appear likely to be resolved through citations of related research. However, concerns regarding reproducibility remain unresolved. Furthermore, the authors’ metrics are somewhat exaggerated to claim they were proposed. Judging solely by this paper, it is difficult to regard it as a complete study.
Review #3
- Please describe the contribution of the paper
This paper proposes a novel framework including a Category-Conditional Anatomical Prototype (CCAP) module and an Uncertainty Aware Decoder for uncertainty-aware landmark localization in ultrasound images. Experiments on a large-scale ultrasound dataset demonstrated the effectiveness of this method and suggest that it may provide an interpretable foundation for ultrasound-guided automated acupuncture.
- 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 motivation of providing an uncertainty-aware probability field instead of a coordinate is reasonable for the task of localizing functional Deqi response zones. The paper proposes a novel CCAP module to retrieve site-specific structural priors and dynamically registers them to the input via deformable alignment. In addition, the experimental results are comprehensive and show that the proposed method outperforms competing approaches in the comparison 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.
1.The EMD term appears to be a supervised Sinkhorn loss between the predicted probability field and a target distribution derived from the ground-truth landmark. However, the manuscript does not explain how the target distribution is constructed from the landmark coordinates, nor whether it is derived from a Gaussian heatmap-based representation or any anatomy-aware supervision. 2.The prototype representation is also insufficiently specified. Important details, such as the size of the prototype and how it is initialized, are missing. 3.The loss formulation is inconsistent across the manuscript. Figure 1 states that L Total = L NLL + L EMD, while Section 2.2 explicitly introduces a L ortho , and Figure 2 further depicts an auxiliary loss. In the implementation details, λ nll, λ sink, and λ reg are reported. The manuscript should present the full training loss explicitly. 4.Several variables are introduced without sufficient explanation, including K, M_k, M_c.
- Please rate the clarity and organization of this paper
Poor
- 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 motivation, model design, and experimental results are interesting and generally convincing. However, the method section lacks sufficient details for reproducibility, which limits the technical clarity of the paper.
- 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.
My main concerns were effectively addressed in the rebuttal.
Author Feedback
We thank the reviewers for recognizing the clinical motivation, CCAP design, and empirical gains. The main concern is reproducibility/clarity (R2,R3), with additional concerns about novelty and scope (R1). We respond based only on the submitted manuscript.
Method/loss reproducibility (R2-W1/4, R3-W1/3/4). We agree the loss should have been stated more explicitly. In Sec.2.1-2.2, the submitted method uses: (i) NLL for the GT coordinate under the predicted heteroscedastic Gaussian field; (ii) Sinkhorn/EMD between P_hat and a normalized target Gaussian centered at y_i at the supervision resolution; and (iii) L_ortho on prototype embeddings. Thus the objective is lambda_nll L_NLL + lambda_sink L_Sinkhorn + lambda_reg L_ortho, with weights in Sec.3.1.The “aux loss” in Fig.2 is the same deep supervision at 1/4 resolution, not another unreported objective. We will revise Fig.1/Fig.2 and Sec.2.2 to use one notation and define K, M_k, and M_c: K is the number of site categories, M_k is the kth learnable prototype, and M_c is selected by the one-hot site indicator c.
Prototype representation (R3-W2). CCAP is not a fixed anatomical atlas. Each prototype is a learnable latent tensor stored in the prototype bank and optimized jointly with the network; it is selected by the site indicator, upsampled to the feature size, warped by the DPA offset field, and injected through gated cross-attention (Sec.2.2, Fig.2). In our implementation, for 384x384 inputs, there is one prototype tensor per site with the same size as the bottleneck feature (C=512, H/16 x W/16), randomly initialized and learned end-to-end. This clarification does not change the model or results.
Nature of uncertainty (R2-W2/3). Our uncertainty is data-dependent aleatoric uncertainty, not epistemic model uncertainty. The variance head predicts sigma^2(x) from each ultrasound image via softplus, and NLL trains the model to relax confidence in ambiguous/deep/shadowed regions (Sec.2.1, Fig.2b). We do not model posterior uncertainty over weights with ensembles or MC dropout. We will state this as “observation-dependent aleatoric uncertainty caused by acoustic attenuation and ambiguous fascial boundaries.”
Evaluation depth and ALHR (R1-W2, R2-W5/6). The study focuses on one clinically defined task: localizing functional Deqi response zones. Within that task, the evaluation covers 1,000 subjects, three anatomically different sites (LI4/LI10/LI8), multi-center acquisition, 5-fold cross-validation, ten baselines, ablation of Proto/DPA/Uncert/EMD, geometric metrics (MRE/SDR), ALHR, and qualitative shadowing/multilayer cases (Sec.3, Tables 1-2, Fig.3). ALHR is not claimed as a general segmentation precision score; it asks whether the predicted landmark peak lies in the expert-defined neuro-fascial layer, complementing MRE/SDR by testing physiological plausibility. We will describe it as a task-specific layer-consistency metric.
Novelty and generalizability (R1-W1, R2 overall). PAPF-Net is more than a loose combination of priors, deformation, and uncertainty: the contribution is the task formulation and integration of (a) site-conditioned latent anatomical prototypes, (b) instance-level deformable alignment, and (c) probabilistic field supervision for functionally verified Deqi labels. The gains in Table 1 and component-wise progression in Table 2 support the structural necessity of these parts. We acknowledge that broader acupoint coverage is future work; the current three sites differ substantially in anatomy and test cross-site heterogeneity rather than a single homogeneous setting.
We will improve clarity and, after de-anonymization if accepted, make the implementation available.
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.
All reviewers are inclined to finalize their decisions following the rebuttal. The authors are advised to address the concerns raised by Reviewers 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 successfully addressed the primary concerns of two reviewers, particularly regarding methodological clarity, uncertainty modeling, and technical validity. The paper presents a clinically motivated integration of several components and is supported by multi-center experiments, strong baselines, and ablation studies. While concerns regarding reproducibility and the completeness of the study remain for one reviewer, the overall evidence suggests that the contribution is technically sound and of interest to the community. Therefore, I recommend acceptance.
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 authors have sufficiently address the concerns of the reviewers. R3 has also moved their score from “weak reject” to “acceptance”.
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.
Following the pre- and post-rebuttal evaluation, my view is that the paper should be accepted. Reviewers 1 and 3 recommend this, and reviewer 2’s remaining criticism does not seem for me to be really unresolved reading carefully the rebuttal. I do agree that the claims of performance are overstated as critisized by reviewer 2, yet this is not a reason for rejection. I do see critical that the code is not released as it reduces the value of the contribution, but that is not a strong argument to push rejection. Overall, I see value to the paper and the work definitely contributes to the community, yet I am not as enthusiastic as reviewers 1 and 3.