List of Papers Browse by Subject Areas Author List
Abstract
Accurate 3D reconstruction of vertebral anatomy from ul-
trasound is important for guiding minimally invasive spine interven-
tions, but it remains challenging due to acoustic shadowing and view-
dependent signal variations. We propose an occupancy-based shape com-
pletion method that reconstructs complete 3D anatomical geometry from
partial ultrasound observations. Crucially for intra-operative applica-
tions, our approach extracts the anatomical surface directly from the
image, avoiding the need for anatomical labels during inference. This
label-free completion relies on a coupled latent space representing both
the image appearance and the underlying anatomical shape. By lever-
aging a Neural Implicit Representation (NIR) that jointly models both
spatial occupancy and acoustic interactions, the method uses acoustic
parameters to become implicitly aware of the unseen regions without
explicit shadowing labels through tracking acoustic signal transmission.
We show that this method outperforms state-of-the-art shape completion
for B-mode ultrasound by 80% in HD95 score. We validate our approach
both in silico and on phantom US images with registered mesh mod-
els from CT labels, demonstrating accurate reconstruction of occluded
anatomy and robust generalization across diverse imaging conditions.
Code and data will be released on publication.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2736_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026/supp/2736_supp.zip
Link to the Code Repository
https://github.com/magdalena-wysocki/Oscar
Link to the Dataset(s)
N/A
BibTex
@InProceedings{WysMag_OSCAR_MICCAI2026,
author = { Wysocki, Magdalena AND Buldu, Kadir Burak AND Gafencu, Miruna-Alexandra AND Killeen, Benjamin D. AND Azampour, Mohammad F. AND Navab, Nassir},
title = { { OSCAR: Occupancy-Based Shape Completion via Acoustic Neural Implicit Representations } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16889},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors propose a physics-aware neural implicit representation for 3D vertebral shape completion from B-mode ultrasound. The core idea is to jointly model acoustic tissue properties and geometric occupancy in a shared latent space, using a differentiable ray-based rendering formulation to simulate acoustic shadowing. At inference, a test-time optimization on raw B-mode intensities, without segmentation labels, infers the complete 3D geometry by exploiting the coupled latent prior. The method targets intra-operative spinal guidance, where acoustic occlusion severely limits direct observation of vertebral anatomy.
- 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.The physics-informed coupling between acoustic properties and geometric occupancy is technically sound and well-motivated.
2.The label-free inference is a genuine practical advantage for intra-operative use, where acquiring segmentation masks at test time is infeasible.
- 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 dataset is very small. Training uses 53 VerSe vertebrae (132 sequences), and quantitative evaluation is performed on 10 simulated and only 3 phantom test subjects. Drawing conclusions about statistical significance from 3 phantom subjects is not credible.
2.On phantom data (Tab. 1), OSCAR achieves HD95 of 7.46 ± 3.41 mm, which is numerically worse than SITD’s 7.17 ± 1.05 mm, and comparable on all other metrics. The authors acknowledge this only with “comparable results,” but the high standard deviation suggests the method is unstable on real data.
3.The comparison with SITD is not on equal footing. SITD requires explicit point cloud labels at inference, while OSCAR does not. Comparing raw metric values between methods with fundamentally different input requirements, without isolating this variable, makes it difficult to attribute performance differences to the proposed technical contributions.
4.There is no ablation study. The paper does not evaluate the contribution of each acoustic parameter individually, nor the impact of the acoustic-geometric coupling versus a geometry-only baseline. It is unclear how much of the gain comes from the physics model versus the occupancy representation alone.
- 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 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
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 proposed method is numerically worse than SITD on HD95 and comparable on all other metrics (Tab. 1). This disconnect between the headline claim and the real-world numbers is the primary concern.
- 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
Review #2
- Please describe the contribution of the paper
This paper propose an occupancy-based shape completion method that reconstructs complete 3D anatomical geometry from partial ultrasound observations. It extracts the anatomical surface directly from the image, avoiding the need for anatomical labels during inference.
- 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.It couples acoustic modeling with geometric occupancy in a shared latent space and leverages differentiable ray-based rendering, enabling shape completion under severe acoustic shadowing. 2.Through test-time optimization, the method reconstructs complete 3D shapes directly from raw B-mode ultrasound images without requiring annotations, showing potential for clinical applications.
- 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 method mainly combines existing NIR, acoustic rendering, and TTO paradigms, and the acoustic module lacks sufficient justification and validation. The claimed “physics-aware” modeling is oversimplified and does not capture realistic ultrasound wave propagation, raising concerns about generalization. 2.The model relies heavily on latent priors to infer occluded regions, rather than leveraging image evidence, which limits its reliability in real-world scenarios. 3.The study lacks real clinical data, the dataset size is very limited, and the experimental evaluation is insufficient, as only two baseline methods are considered. 4.The contribution of each component (acoustic modeling, latent coupling, TTO) is not clearly validated. 5.Hyperparameters are not described in sufficient detail, and the phantom experiments are statistically insignificant.
- 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.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
1.Lack of motivation. There are no clear technical difficulties to deal with. 2.The design of each module adopts backward technology and wrong description. 3.The experimental evaluation is insufficient.
- 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
Review #3
- Please describe the contribution of the paper
The authors propose OSCAR, an occupancy-based model for 3D shape completion in spine ultrasound that jointly represents the acoustic signature (acoustic reflection, scattering and attentuation) and geometric (volume occupancy) anatomical properties. The proposed model’s latent space has a bidirectional nature which also supports synthesis of novel views of B-mode images when the shape is known. Modelling the acoustic properties allows the model to reconstruct even unobserved regions with existing information. It yields an 80% improvement in HD95 over the SOTA (SITD). OSCAR has been validated on both in silico and phantom US images with vertebrae models obtained from the opensource VerSe dataset which confirms that it can be translated to real-world settings successfully.
- 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-structured and written very clearly with appropriate visualisations. The formulations are concise and clear which improves the readability of the paper.
- It is interesting to see the joint feature representation of the vertebra taking into account both the physics-based properties and the shape and morphology for shape completion.
- The visual depictions of the results in the Evaluations section, especially the demonstrations of the bidrectional nature of the shared latent space and anatomically plausible intermediate states in latent interpolation is commendable and explains the methodology very well.
- Clinical validation in both simulation and phantom settings highlights its promise for real-world applications.
- 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.
In Eq. 3, the authors have not justified how they arrived at the corresponding weights for each of the Loss component. Similarly, the chosen value for alpha in Eq. 4 is also not mentioned. The paper would also be strengthened by also reporting the inference time taken as the authors have performed preliminary clinical validation both on simulation and phantom to show that the method is applicable to real-world scans.
- 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 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
There appears to be a notation inconsistency in Fig. 1, where the s_psi is labeled as geometric head, while it is referred to as occupancy head everywhere else in the paper. In Fig 5, the images used for OSCAR seem to be stretched horizontally.
- 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.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper is very well-structured and clearly written. The visualizations in the Evaluation section effectively aid both experts and readers with a background in medical image analysis in understanding the methodology and results. The proposed shape completion approach can be potentially extended to other ultrasound imaging datasets as well as it holistically takes into account the shape and physics-based properties of the anatomy. Therefore, I believe this paper merits acceptance.
- 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.
I believe the authors have satisfactorily addressed the reviewers’ concerns and provided well-reasoned explanations. The proposed idea could also be extended to broader range of ultrasound applications. Overall, this work presents an interesting and significant contribution to the field.
Author Feedback
We thank reviewers for their feedback. Reviewers recognized the novelty of acoustic-geometric coupling (R1,R2,R3), its potential for label-free intraoperative guidance (R1,R2), and the bidirectional latent space enabling image synthesis (R3). Dataset size (R1,R2): We agree that larger evaluation would be needed to determine the clinical applicability of OSCAR. We believe the evaluation here (on 10 simulation subjects with varying B-mode and 3 physical phantoms with unseen anatomy in tissue-mimicking gelatin) demonstrates generalizability sufficient to motivate ground truth annotation, ethical approval, and organization in the context of costly CAI data collection. We will update the experimental details to highlight comparable phantom evaluations in prior work [2,16] and the discussion to outline future studies. Variance on phantom validation (R1): Our phantom experiments specifically show the generalizability of OSCAR trained on simulation to the real B-mode images. The variance in the results relate to the domain gap. This gap can be bridged by improving the fidelity of the simulation. We will clarify this point in the discussion section. Acoustic model ablation (R1,R2): While we agree that isolating the contribution of each component may improve understanding of this novel class of models, we believe OSCAR represents new capabilities, namely (a) label-free US-based shape completion and (b) image synthesis contingent on (b). It adopts the forward synthesis model from prior work [1,11,15] with scattering σ, attenuation μ, and reflection β, improving shape completion accuracy with the intermediate step between B-mode and MLP. Removing each parameter shows β is responsible for transmission and received echo at tissue boundaries. β does not reconstruct B-mode faithfully, losing bidirectional coupling and image synthesis. Removing μ is partially compensated by β, but μ must be coupled with σ since μ does not form the image. We will update the method to emphasize these points and point toward the opportunity for comprehensive analysis beyond the scope of this initial work. Geometry-only ablation (R1,R2): We will update the CR version to clarify that NISF results represent this ablation since it features the same backbone but with no acoustic features. Likewise, decoupling geometry and acoustics with two separate networks sharing a latent vector gives worse performance (HD95 1.17 to 3.23 mm), motivating the joint architecture. TTO ablation (R2): The number of val iterations is fixed during training. TTO starts from the mean shape (avg HD95 5.19 mm) and improves until the final iteration. SITD Comparison (R1,R2): We thank reviewers for raising an important distinction between SITD and OSCAR: SITD only achieves comparable performance because it has access to bone segmentation annotations. OSCAR is label-free, which R1 recognizes as practically advantageous because “acquiring segmentation masks at test time is infeasible.” We will update Tab 1 to clarify this and the discussion to highlight the need for benchmarks on label-free methods. Physics simplification (R2): We agree that full wave-equation modeling is more physically faithful. However, for differentiability and computational feasibility, we adopt a simplified model widely used in ray-based differentiable US simulation [1,10,15], which we will clarify in the introduction. Shape priors (R2): We will clarify that during TTO, shape completion starts from the prior distribution and updates the latent code to match the B-mode. Where image evidence exists, the transmission-gated loss drives the latent code update; shadowed regions contain no signal by definition, so the prior completes the anatomy, governing what is acoustically unobservable. Minor comments (R2,R3): We will motivate all hyperparameter values and correct figs in CR. OSCAR offers INRs coupling acoustic and anatomical spaces, enabling label-free shape completion and bidirectional image-shape synthesis for intra-op US.
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 for this are divergent, ranging from Accept to Reject. All reviewers recognized the novelty of coupling acoustic properties with geometric occupancy, and its potential for clinical applications such as intra-operative spinal guidance. The label-free inference mechanism is recognized as the key contribution and offers significant practical advantage for real-world use where annotations are unavailable. Manuscript is well-written and the high-quality visualizations. Despite this, reviewers point out several concerns with quantitative analysis, with a key concern being the size of dataset (10 simulation), and the lack of ablation studies to better attribute the contribution of various components (acoustic modeling, latent coupling, TTO). R1 also raises concerns with high variance of metrics such as HD and CD. I believe the paper has strong merits and many of these concerns are potentially addressable via rebuttal and thus encourage authors to provide a response addressing reviewers’ concerns.
- 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 initially expressed concerns regarding the limited dataset size, lack of ablation studies to help justify the proposed approach as well as high variance in HD and CS metrics. The rebuttal addressed these concerns by clarifying the experiment setup, emphasizing the label-free nature of the method as key advantage, and providing detailed explanation regarding contributions of individual components. R1 still maintains rejection, noting the SITD outperforming OSCAR on real data on HD; however, authors clarified that SITD relies on label information and performance is comparable (SITD: 7.17 ± 1.05 vs OSCAR: 7.46 ± 3.41). Although OSCAR builds existing techniques, the coupling of acoustic and geometric information within a unified latent space, and feasibility of label-free shape completion has merits. Thus, AC recommends acceptance. However, authors must update the manuscript to address the reviewers’ concerns, particularly addressing the contribution of individual components as well as suggestions on addressing the simulation fidelity (as noted in rebuttal).
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Reject
- Please justify your recommendation.
The rebuttal reframed the SITD gap as acceptable because OSCAR is label-free and described ablation outcomes, but it conceded the larger evaluation needed to substantiate the method and introduced these ablations as future camera-ready additions rather than evidence in the submitted work, so the central contribution remains unvalidated on real data. Therefore, the recommendation is Reject.
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 work proposes a physics-aware neural implicit representation for 3D vertebral shape completion from B-mode ultrasound, with label-free inference and a coupled acoustic-geometric latent representation that is relevant for intra-operative spinal guidance. The authors satisfactorily addressed the reviewer’s concerns and provided well-reasoned explanations. The approach is technically interesting, has potential applicability beyond this specific vertebral ultrasound setting, and represents a meaningful contribution to ultrasound-based shape completion.