Abstract

Intraoperative Cone Beam Computed Tomography (CBCT) provides a reliable 3D anatomical context essential for interventional planning. However, its static nature fails to provide continuous monitoring of soft-tissue deformations induced by respiration, probe pressure, and surgical manipulation, leading to navigation discrepancies. We propose a deformation-aware CBCT updating framework that leverages robotic ultrasound as a dynamic proxy to infer tissue motion and update static CBCT slices in real time. Starting from calibration-initialized alignment with linear correlation of linear combination (LC2)-based rigid refinement, our method establishes accurate multimodal correspondence. To capture intraoperative dynamics, we introduce the ultrasound correlation UNet (USCorUNet), a lightweight network trained with optical flow-guided supervision to learn deformation-aware correlation representations, enabling accurate, real-time dense deformation field estimation from ultrasound streams. The inferred deformation is spatially regularized and transferred to the CBCT reference to produce deformation-consistent visualizations without repeated radiation exposure. We validate the proposed approach through deformation estimation and ultrasound-guided CBCT updating experiments. Results demonstrate real-time end-to-end CBCT slice updating and physically plausible deformation estimation, enabling dynamic refinement of static CBCT guidance during robotic ultrasound-assisted interventions. The source code is publicly available at https://github.com/ziyuan-li/robotic-us-cbct-updating.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/5031_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: https://papers.miccai.org/miccai-2026/supp/5031_supp.zip

Link to the Code Repository

https://github.com/ziyuan-li/robotic-us-cbct-updating

Link to the Dataset(s)

N/A

BibTex

@InProceedings{LiFen_Robotic_MICCAI2026,
        author = { Li, Feng AND Li, Ziyuan AND Jiang, Zhongliang AND Navab, Nassir AND Bi, Yuan},
        title = { { Robotic Ultrasound Makes CBCT Alive } },
        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

    This paper presents a deformation-aware updating framework for static Cone Beam Computed Tomography (CBCT) images using real-time robotic ultrasound (US). To address the critical issue of intraoperative soft tissue deformation that compromises surgical navigation, the authors propose USCorUNet. This network estimates the real-time deformation field directly from the US stream, allowing for the dynamic updating of preoperative/intraoperative CBCT slices without subjecting the patient to additional radiation.

  • 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 introduction of the USCorUNet to continuously estimate deformation fields from US streams is technically sound. It elegantly bridges the gap between the high refresh rate of US and the static nature of CBCT. The integration of calibrated alignment with deep learning-based deformation estimation demonstrates a comprehensive understanding of robotic ultrasound-guided surgical systems. The achieved real-time performance is crucial for intraoperative deployment.

  • 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.

    Ultrasound imaging is notoriously susceptible to artifacts such as acoustic shadowing, speckle noise, and signal dropouts. The paper lacks a detailed discussion or specific experiments on how USCorUNet handles sudden drops in US image quality. The description of the validation dataset in the current manuscript could be more detailed. It is not entirely clear to what extent the framework was validated on highly realistic in-vivo data versus idealized phantoms. Phantom studies, while valuable for baseline metrics, often fail to replicate the complex non-rigid deformations and heterogeneous textures of real human tissue.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission has provided an anonymized link to the source code, dataset, or any other dependencies.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (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?

    This is a solid paper with a compelling clinical application and a well-designed technical pipeline. I am inclined to recommend acceptance, provided the authors can adequately address the concerns regarding robustness to US artifacts and clarify the validation dataset details during the rebuttal phase.

  • 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 have no more comments.



Review #2

  • Please describe the contribution of the paper

    The paper presents a deformation-aware CBCT updating framework that uses robotic ultrasound as a dynamic proxy. The main contributions are: (1) an end-to-end workflow from calibration/registration to real-time CBCT slice updating, (2) a lightweight network USCorUNet that estimates dense deformation fields from adjacent ultrasound frames without ground-truth deformation labels, leveraging RAFT-derived pseudo-labels combined with self-supervised photometric and regularization losses, and (3) a practical system integrating a KUKA robot, Siemens ultrasound, and CBCT, demonstrated on phantoms and in vivo arm data.

  • 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 authors provide a full pipeline from hand-eye calibration, LC2-based rigid refinement, deformation estimation, to CBCT slice updating. This is a valuable engineering contribution for robotic ultrasound-guided interventions.

    2.Since ground-truth deformation fields are unavailable for ultrasound, the method distills pseudo-labels from RAFT optical flow, then refines them with confidence-weighted photometric loss and Jacobian-based folding regularization. This semi-supervised + self-supervised approach is sensible and well-motivated.

    3.USCorUNet runs at ~11ms end-to-end, enabling real-time intraoperative use. The local correlation volume and 1/8-resolution decoding strike a reasonable trade-off between accuracy and speed.

  • 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 authors state that LC2 searches “within a constrained range” after calibration initialization, but no numerical values (e.g., translation/rotation limits) or validation of convergence range are provided.

    2.Although a Gaussian correction profile is introduced for probe compression, the paper does not quantify the relationship between applied force and deformation estimation error. The model assumes a symmetric convex profile, which may not hold for heterogeneous tissues or off-center indentation.

    3.All metrics are indirect (photometric consistency, forward-backward error, folding ratio, bone mask Dice). No validation against physical markers, finite-element simulations, or ex vivo tracked targets is provided. The clinical plausibility of the estimated deformation fields still remains partially unverified.

    4.If I understand correctly, the method updates only the single CBCT slice coplanar with the ultrasound plane, not the full 3D volume. This restricts applicability for true 3D navigation.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission has provided an anonymized link to the source code, dataset, or any other dependencies.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (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 presents a complete, real-time deformation-aware CBCT updating framework with a lightweight network trained without ground-truth deformation labels, achieving strong results on phantoms and in vivo data. Limitations include unclear capture range, indirect validation, and limited clinical scenarios, but these do not detract from the clear technical contribution and reproducibility.

  • 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 still hold the same view that the paper can be accepted.



Review #3

  • Please describe the contribution of the paper

    The paper proposes a workflow for updating CBCT images with live ultrasound during image-guided interventions. The framework includes ultrasound calibration, registration, and CBCT deformation, and introduces USCorUNet for ultrasound-to-ultrasound deformation field estimation. Overall, the topic is highly relevant to the MICCAI community, as it addresses an important challenge in image-guided intervention and aims to improve the integration of complementary intraoperative imaging modalities.

  • 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 manuscript has several strengths.

    • It presents a complete workflow rather than an isolated methodological component, which enhances the translational value of the work and makes the proposed approach more compelling from an application perspective.

    • The idea of leveraging a RAFT-inspired optical flow strategy to capture deformations between consecutive ultrasound frames is interesting and well aligned with the dynamic nature of deformation induced in ultrasound imaging.

    • The validation is also supported by a relatively diverse dataset, including phantom data, and the comparison against other U-Net-based approaches is appreciated. These elements together make the study relevant and potentially impactful.

  • 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 methodological contribution would benefit from clearer positioning. Several of the key building blocks of the framework appear related to existing work, including LC2-based multimodal registration, deformable registration approaches, and CBCT updating strategies. My understanding is that the contribution is focused on USCorUNet. As currently presented, the statement of contributions is somewhat misleading. A clearer articulation of what is fundamentally novel in the proposed framework would help the reader better understand the scope.

    • The central problem appears to be the ultrasound-to-ultrasound deformation estimation. However, it is not yet clear how the proposed solution compares with more classical frame-to-frame ultrasound registration strategies, such as a simple NCC-based approaches. This comparison seems particularly important, as it would help establish rationale to whether the proposed method provides a meaningful advantage over simpler and more established alternatives.

    • Another point that would benefit from clarification is how out-of-plane deformation is handled. In addition, the video examples appear to show some discrepancy in the out-of-field ultrasound deformation. It would therefore be helpful for the manuscript to discuss how such deformations are modeled, and more generally, how robust the method is when out-of-plane or out-of-field motion occurs.

    • There are also some concerns regarding the experimental setup and evaluation. The study mentions scans of in vivo forearm and upper-arm ultrasound (Dataset A), but no ethics approval or related statement appears to be provided.

    • The results presented in Table 2 are difficult to interpret without a clear understanding of the force-stratified evaluation protocol. A concise explanation of this protocol would greatly improve the readability and interpretability of the results.

    • The fine-tuning strategy also raises some questions. The manuscript indicates that fine-tuning was performed on different datasets while keeping the train, validation, and test ratios unchanged. As presented, Table 3 seems to report mean results on test sets drawn from datasets that were also used during training or adaptation. This makes it difficult to judge whether the reported improvements reflect true generalization or are partly driven by dataset-specific adaptation. Clarifying this point would be important, particularly for understanding how the method is expected to perform in clinical practice, where the deformation source is not known in advance. Also a discussion of the anticipated robustness and expected performance under such realistic conditions would substantially strengthen the work.

  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission has provided an anonymized link to the source code, dataset, or any other dependencies.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    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?

    Overall, this is a relevant and promising paper that addresses an important translational problem. The proposed workflow is comprehensive, and the technical implementation appears strong. The manuscript would be significantly strengthened by a clearer and more precise articulation of its methodological novelty, as the current framing tends to overstate the contributions. In particular, stronger justification for the ultrasound-to-ultrasound registration approach, along with a clearer description of the evaluation and its implications for clinical generalization, would help better support the paper’s claims and impact.

  • 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.

    The authors have addressed most of my comments. I recommend acceptance, provided that the justifications and discussions included in the response are incorporated into the final version of the paper.



Author Feedback

We thank the reviewers and AC for their careful assessment and summary. We appreciate R1/R2’s encouraging evaluations (recommended Accept) of our full pipeline, USCorUNet design, and accuracy–speed trade-off, R3’s constructive suggestions and recognition of the complete workflow, RAFT-inspired deformation modeling, diverse validation, and translational value, as well as the AC’s summary.

Methodology (R3): The reviews highlight our core contributions: (i) a pipeline from hand-eye calibration and LC2 refinement to deformation estimation and CBCT slice updating; (ii) USCorUNet, a lightweight bidirectional network estimating dense deformation without ground-truth labels via RAFT-derived pseudo-labels and semi-/self-supervised losses, running at ~11 ms; (iii) real-time CBCT updating for probe- and externally induced deformations, bridging high-refresh-rate US and static CBCT; and (iv) validation on in vivo and three phantom datasets. As suggested, we will clarify the distinction between established building blocks and our task-specific integration, USCorUNet, and CBCT-updating contributions.

Baselines (R3): NCC/block matching (BM) is a useful local patch-tracking reference, but mainly measures local similarity rather than dense, physically plausible deformation. Prior works (Che et al., “Ultrasound Registration …”, Methods 2017; Sotiras et al., “Deformable Medical …”, TMI 2013; Alessandrini et al., “Detailed Evaluation …”, TMI 2016) show that US registration is sensitive to noise/artifacts/decorrelation, and BM can have larger bias/errors than flow-based methods. Thus, we used task-level baselines: RAFT [18] for dense US-US motion, DefCor-Net [9] for force-stratified comparison, and LC2-FFD [4] for CBCT updating. Together with forward-backward (FB) consistency, folding ratio, Dice, and CBCT-update metrics, they assess motion quality, anatomical/physical plausibility, and downstream utility beyond patch similarity.

Validation metrics (R1/R2): We use MAE/NCC for alignment, FB residual for inverse consistency, folding ratio for physical plausibility, bone-mask Dice for anatomical marker correctness, and CBCT-update metrics. For shadowing/dropout/speckle decorrelation, the confidence-weighted photometric loss [10] downweights unreliable regions, while Jacobian regularization suppresses implausible folding; Table 1 ablations show worse FB/Dice/folding scores when either confidence or regularization is removed. We agree that FEM/ex vivo tracked targets would further strengthen validation and will state this point; our current evidence combines multi-dataset evaluation with task-level CBCT updating.

Scope/out-of-plane motion (R2/R3): The method updates the 2D CBCT slice corresponding to the active 2D US plane; full 3D volume updating is not claimed. Out-of-plane motion is not directly observed by 2D US. Thus, EDT/Gaussian spatial weighting lets deformation decay away from the observed US window, rather than inferring out-of-plane deformation without additional image evidence.

Dataset/protocol (R1/R3): Dataset A uses data from DefCor-Net [9], which was collected under IRB approval No. 2022-87-S-KK. Table 2 follows the force-stratified protocol of DefCor-Net: force between robot and probe is synchronized with B-mode frames and grouped into 1–6 N. Table 3 reports regime-level adaptation, not per-test fitting: the probe-adapted model uses only training splits pooled from Datasets B–D, and the external-adapted model only uses the Dataset B training split; all test splits are held out.

LC2/Gaussian details (R2): LC2 searches in-plane translations within ±10 CBCT pixels (0.655 mm/pixel) around the projected US position; slice/orientation are fixed by calibrated geometry. The Gaussian profile is a spatial prior for transferring local US motion to the CBCT ROI, not a calibrated force-to-error model. We will clarify this in the camera-ready version.

We appreciate all constructive feedback and recognition of our contributions.




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.

    This paper presents a clinically relevant framework for deformation-aware CBCT updating using real-time robotic ultrasound, integrating calibration, registration, deformation estimation, and CBCT slice updating into a complete workflow. Reviewers generally agree that the system-level contribution is strong, the real-time performance is promising, and USCorUNet is a technically sound component for ultrasound-based deformation estimation without ground-truth deformation labels. However, concerns remain regarding the precise methodological novelty, the lack of comparison with simpler ultrasound registration baselines, the use of indirect validation metrics, robustness to ultrasound artifacts and out-of-plane motion, and the clarity of dataset/evaluation protocols.

  • 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 effectively addressed the reviewers’ concerns, providing sufficient clarification and additional evidence where requested. Following the rebuttal, all reviewers expressed support for acceptance, indicating a clear consensus that the paper meets the standard for acceptance at MICCAI.



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 main concerns (R3) appear resolved after rebuttal, namely clarification of novelty and methodological details. The authors are encouraged to incorporate rebuttal clarifications in the camera ready version of their paper.

    I recommend this submission is accepted



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.

    All 3 reviewers are overall satisfied with the rebuttal and recommend acceptance.

    The authors have done a good job in the rebuttal clarifying all criticism. However, since it is not explicitly stated the authors must incorporate all major points in their rebuttal in the final manuscript.

    I recommend acceptance.



back to top