List of Papers Browse by Subject Areas Author List
Abstract
We introduce BridgeSplat, a novel approach for deformable surgical navigation that couples intraoperative 3D reconstruction with preoperative CT data to bridge the gap between surgical video and volumetric patient data. Our method rigs 3D Gaussians to a CT mesh, enabling joint optimization of Gaussian parameters and mesh deformation through photometric supervision. By parametrizing each Gaussian relative to its parent mesh triangle, we enforce alignment between Gaussians and mesh and obtain deformations that can be propagated back to update the CT. We demonstrate BridgeSplat’s effectiveness on visceral pig surgeries and synthetic data of a human liver under simulation, showing sensible deformations of the preoperative CT on monocular RGB data. Code, data, and additional resources can be found at https://maxfehrentz.github.io/ct-informed-splatting/.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/4699_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2025/supp/4699_supp.zip
Link to the Code Repository
https://maxfehrentz.github.io/ct-informed-splatting/
Link to the Dataset(s)
https://maxfehrentz.github.io/ct-informed-splatting/
BibTex
@InProceedings{FehMax_BridgeSplat_MICCAI2025,
author = { Fehrentz, Maximilian AND Winkler, Alexander AND Heiliger, Thomas AND Haouchine, Nazim AND Heiliger, Christian AND Navab, Nassir},
title = { { BridgeSplat: Bidirectionally Coupled CT and Non-Rigid Gaussian Splatting for Deformable Intraoperative Surgical Navigation } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15970},
month = {September},
page = {44 -- 53}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper introduces BridgeSplat, an approach for deformable surgical navigation that couples intraoperative 3D reconstruction with preoperative CT data to bridge the gap between surgical video and volumetric patient 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.
This paper provides video demonstrations of simulation results, which are of certain clinical significance. Additionally, the method of combining Gaussian splatting with preoperative CT data to achieve dynamic deformation of the CT also has a certain clinical value.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1）This paper has significant structural issues. It lacks comparison study and sufficient demonstration of results across multiple datasets, making it insufficient to showcase the superiority of the proposed algorithm. 2）This paper is poorly formatted, with matrix sets in paragraphs lacking variable names. 3）The Conclusion states that the paper integrates 4D Gaussian splatting with preoperative CT. However, the Method and Experiments and Results sections use 3D Gaussian splatting without explaining the connection between them.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- 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/2025/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 — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper’s conclusion doesn’t align well with the methodology, i.e., while the conclusion claims integration of 4D Gaussian splatting with preoperative CT, the methods and experiments sections actually use 3D Gaussian splatting, without clarifying the relationship between them. Additionally, the paper architecture is confused, and has formatting issues, leaving much room for improvements.
- 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 addressed most of my concerns. I recommend the acceptance.
Review #2
- Please describe the contribution of the paper
- This paper proposes a novel coupling of 3D Gaussian Splatting with a deformable CT-derived mesh, enabling intraoperative reconstructions to propagate mesh deformations back to the CT scan.
- The authors introduce a constrained version of Gaussian Splatting that anchors Gaussians to the CT-derived surface mesh, ensuring that the reconstruction remains consistent with the mesh geometry.
- The authors validate their approach both quantitatively on synthetic liver simulations and qualitatively on real pig surgery videos.
- 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.
- Level of Novelty The main workflow of bidirectional coupling between CT and intraoperative monocular views is a novel contribution. The methodology discussed in section 2.2, 2.3 and 2.4 are based on reference [11], [6], and [15] with incremental modifications to make the Gaussians conforming to the mesh.
- Quantitative and Qualitative Validation The authors have evaluated their approach on liver deformation simulations and two visceral pig surgeries, demonstrating both numerical accuracy (within 5 mm) and visually plausible deformation tracking.
- Practicality and Simplicity The method works with monocular 2D RGB video only, reducing hardware requirements for surgical 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.
- Limitations on Methodology 1) The method requires accurate initial rigid registration between the CT and intraoperative scene. 2) The method uses first frame visible Gaussians as anchor points which may not be perfect for long sequences or large camera movement. 3) Deformation modeling is limited on mesh surface, which can be inrealistic for certain large or internal deformations.
- Error Analysis and Validation The authors report absolute values on Euclidean distance for error analysis, without providing any relative errors based on deformation amplitude or total strain. Also, no direct comparisons with other 4DGS-based or SLAM-based methods are provided.
- Lack of Real-Time Performance The authors do not provide any numbers on computational efficiency of their method, mentioning their method does not work in real time yet.
- 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.
- 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/2025/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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Despite its multiple limitations, I appreciate the novelty and simplicity of the proposed method.
- Reviewer confidence
Somewhat confident (2)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
I appreciate the authors’ response and clarification regarding the lack of comparisons with prior methods such as SLAM-based methods. Given its limitations, I still appreciate the novelty and simplicity in the methodology.
Review #3
- Please describe the contribution of the paper
The authors proposed a novel approach for deformable surgical navigation that performs intraoperative 3D reconstruction with preoperative CT data to bridge the gap between surgical video and volumetric patient 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.
Novel method for performing the deformable intraoperative registration by using GaussianSplatting.
- 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.
-
“However, it also comes with challenges, as tactile feedback is limited, spatial understanding is demanding, handling is challenging, and learning curves are flat.” This sentence lacks citation.
-
I understand the authors treat the intraoperative registration as a camera tracking task. However, how did the authors determine the first frame camera pose?
-
why the authors did not include the rotational error analysis?
-
Is there any failure cases? Could the authors discuss on those in the paper?
-
- 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.
- 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/2025/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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Overall, it is a novel paper that tackles a long-standing challenging research task of performing intraoperative registration of deformable objects in surgery. Although cannot be put in use clinlically yet due to the method is not real-time; however, the whole idea of this project seems promising.
- 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.
This paper is unable to provide rotational error analysis. For surgical navigation, it is consider an important practice to include some analysis. There are also ways to determine the first camera pose, such as directly registering 3D data to 2D scene to obtain the estimated first camera pose in videos. In real surgery setting, the camera poses are unlikely to be given in real time, I have questions in the possible real-world application of such deformation-focused method.
Most importantly, this paper lacks sufficient quantitative comparison and heavily rely on the qualitative results, which raise concerns about the credibility of the final results as it can introduce bias as they can simply give better visual results but ignore the worse performing ones.
There are also no comparing methods to help the reviewer judge the merit of the methodology contributions.
Overall, I would suggest to reject this paper.
Author Feedback
We are pleased that the reviewers recognize the merit of our method and provided valuable and actionable feedback on the paper. R3 and R4 highlighted the novelty of coupling a 4D radiance field with preoperative CT. R3 notes the practicality and simplicity of the method. R1 appreciated the clinical significance of deforming the preoperative CT from intraoperative video. We value their constructive feedback and address the major comments below.
UNCLEAR RELATIONSHIP BETWEEN 3DGS AND PREOPERATIVE CT (R1): R1 mentions that method, experiments, and results use 3DGS without explaining the connection between Gaussian Splatting and preoperative CT. We tried to highlight the connection in Fig. 1 and explained it in Sec. 2.1. We obtain a preoperative mesh M of the abdominal cavity from the preoperative CT. The 3D Gaussians are then coupled to the mesh M as further explained in section 2.2 “Coupling 3D Gaussians and Mesh”. We refer to this as 4D Gaussian Splatting in the paper, as commonly done when describing any variation of dynamic 3D Gaussians. We will make sure to clarify this in the final version.
EVALUATION & COMPARISON AGAINST OTHER METHODS (R1, R3): Although R3 highlighted that we provide qualitative and quantitative results on clinical and simulated data, respectively, both R1 and R3 rightfully point out that a larger-scale evaluation on multiple datasets and a comparison against other methods is desirable. Constraints on data: Our method requires intraoperative video and preoperative CT. Although there are multiple well-known open-source datasets for MIS (e.g., DePoLL, EndoNeRF, …), unfortunately, none of them come with the corresponding preoperative CTs and camera poses. Therefore, the only data suitable for the task was the one we collected after major efforts. Collecting more data is a natural step for future work. Constraints on comparison against other methods: Other 4D radiance field methods for MIS rely solely on RGB videos without 3D priors, whereas our approach is explicitly coupled to one. We opted against direct comparisons for two reasons: (a) our CT-based prior would give us an unfair advantage, and (b) no meaningful ground truth or metric exists — clinical data lacks 4D ground truth, and photometric reconstruction quality (PSNR, …) is not the right measure based on our objective. For comparison on simulated data, we could have initialized the other 4DGS methods with a point cloud from the mesh. However, other MIS 4DGS methods lack surface guarantees. In contrast, our mesh-coupled representation provides an inductive bias, making surface comparisons inherently biased in our favor.
QUESTIONS ON CAMERA TRACKING/SLAM (R3, R4): We want to clarify that our method does not estimate camera poses. R3 requested a comparison against SLAM-based methods, and R4 asks how we determine the first camera pose and wishes to see rotation error analysis. Please note that this is the first work that aims at deformation estimation from tracked monocular 2D video only, based on a preoperative CT. Tracking the camera would require a deformable SLAM method, which is a highly ambiguous and unsolved problem, especially from monocular input. As noted in the Introduction: “Despite all the progress on SLAM, as concluded in [13], tracking of the camera is still desirable, if not necessary, to move towards clinically feasible systems.“ Therefore, we use a tracked laparoscope to focus the problem on deformation tracking, assuming an OR setup that is already clinical practice in neurosurgery.
OTHER REQUESTED MINOR IMPROVEMENTS IN CAMERA-READY VERSION (R1, R3, R4): We will follow R1 and revise the formatting. As requested by R3, we will add a brief discussion of computational demands and runtime. Following R4, we will add a brief discussion of failure modes. To ensure reproducibility (R1), as mentioned in the Abstract, we will make our data and code public upon acceptance.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #3
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
The rebuttal addresses most of the concerns, but still leaves some open questions/concerns, esp. the paper lacks rotational error analysis, which is crucial for surgical navigation. It also relies too heavily on qualitative results without sufficient quantitative comparison or baseline methods, undermining the credibility and evaluation of its contributions. So, this paper is not ready to be accepted at MICCAI’25.