List of Papers Browse by Subject Areas Author List
Abstract
Surgical scene reconstruction from endoscopic video is crucial for many applications in computer- and robot-assisted surgery. However, existing methods primarily focus on soft tissue deformation while often neglecting the dynamic motion of surgical tools, limiting the completeness of the reconstructed scene. To bridge the aforementioned research gap, we propose T^2GS, a novel and efficient surgical scene reconstruction framework that enables efficient spatio-temporal modelling of both deformable tissues and dynamically interacting surgical tools. T^2GS leverages Gaussian Splatting for dynamic scene reconstruction, and it integrates a recent tissue deformation modelling technique while most importantly, introduces a novel efficient tool motion model (ETMM). At its core, ETMM disambiguates the modelling process of tool’s motion as global trajectory modelling and local shape-change modelling. We additionally propose pose-informed pointcloud fusion (PIPF), holistically initialized of tools’ gaussians for improved tool motion modelling. Extensive experiments on public datasets demonstrate T^2GS’s superior performance for comprehensive endoscopic scene reconstruction compared to previous methods. Moreover, as we specifically design our method with efficiency in concern, T^2GS also showcases promising reconstruction efficiency (3mins) and rendering speed (71fps), highlighting its potential for intraoperative applications. Our code is available at https://gitlab.com/nct_tso_public/ttgs.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/5019_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{XuJin_T2GS_MICCAI2025,
author = { Xu, Jinjing and Li, Chenyang and Liu, Peng and Pfeiffer, Micha and Liu, Liwen and Docea, Reuben and Wagner, Martin and Speidel, Stefanie},
title = { { T2GS: Comprehensive Reconstruction of Dynamic Surgical Scenes with Gaussian Splatting } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15968},
month = {September},
}
Reviews
Review #1
- Please describe the contribution of the paper
A 4DGS-based framework is designed for holistically reconstruct surgical scenes, which is featured by its tool motion reconstruction module ETMM and tools’ Gaussians initilization method PIPF.
- 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 key innovations include the proposed Efficient Tool Motion Modelling (ETMM) module and pose-informed pointcloud fusion (PIPF) method. The former disambiguates the modelling process of tool’s motion as global trajectory modelling and local shape-change modelling. The latter holistically initialized of tools’ gaussians to improve the tool motion reconstruction performance. The resulting framework achieves efficient and high-quality reconstruction of comprehensive surgical scenes, particularly in tool reconstruction quality.
- 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) There are obvious mistakes in Table 1. The data items (36.53/ 35.91), (95.04/ 94.69), (34.08/ 33.64) and (92.25/ 91.60) are misplaced. The training time and rendering speed of T2GS for moving tool reconstruction are not reported. The T2GS’s SSIM score 92.12 is not the best one but the second best one, so it should be shown with underline. Besides, what is the intention to report the T2GS’s scores in the tissue-only areas? 2) The paper only has one figure in the method part. However, this figure is not informative and hardly helps the readers to understand the framework. For example, what do “L_color” and “L_depth” mean? They are not mentioned in the text. What specific messages does Fig. 1(b) intend to convey?
Other mistakes: Page 3: “Sec ??”. Page 3: “where α′” should be “where α′_j” Page 4: “2D data-association of 2D points p^t+1_i and p^t+1_i”. The first “p^t+1_i” should be “p^t_i”? Page 4: Please check “Meaning of global motions and local shape changes… Therefore, we design the two modules … and … to model” Table 1: “stared” should be “starred”
- 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.
- 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 is well motivated and the proposed framework demonstrates the improved tool reconstruction performance and good training/rendering efficiencies. However, there are some obvious mistakes in the current version. The figure quality should be improved. The reconstruction performances of the proposed method is not significantly better than SurgicalGS, despite the training time.
- 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 targets at dynamic surgical scene reconstruction with both deformable tissues and dynamically interacting surgical tools. To achieve that, the paper proposes a new efficient tool motion module to estimate the motion of operating tools based on poses. They additionally initialize the tools’ gaussians with pose-informed pointcloud fusion. Experiments on two datasets demonstrate the effectiveness and efficiency of the proposed method.
- 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 research gap of modelling the dynamic motion of both the deformable tissues and interacting surgical tools is interesting.
++ The proposed tool motion module, which includes the Pose-guided Global Trajectory Modelling and Local shape-change Modelling are, is new and reasonable for handling tool’s motion.
- 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 Table 1, the paper compares with other methods on complete frame and tool-area only regions. However, since previous methods are not designed to handle tool motion, it is suggested to also include tissue-only comparisons with previous methods to demonstrate the advancement of the proposed method on all dynamic contents.
– In Table 1 Moving Tool Reconstruction, why the proposed method requires “95.04”, which is much longer than other approaches, such as “1:28” from Deform3DGS. Is this reported metric correct?
– The proposed method optimizes the tool poses in Sec. 2.3. However, how to evaluate the accuracy of estimated tool poses is unclear. It would be better to conduct evaluation for estimated tool poses as well.
– Section 2.4 is incomplete with much “…” in the first paragraph.
– Typo: Sec. ?? in Page 3. Two same p_i^t+1 in page 4.
- 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.
- 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?
The paper’s contributions on dynamic surgical reconstruction with both deformable tissues and operation tools are interesting. The proposed designed modules are new and interesting. Despite there are some typos and concerns on the experiments part, the paper’s methodology part is reasonable. Therefore, the reviewer is leaning towards weak accept and suggests the authors to response clearly on the weaknesses part.
- 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 core contribution lies in the design of the Efficient Tool Motion Modeling (ETMM) module, which decomposes tool motion into global trajectory modeling (P-GTM) and local shape-change modeling (LSM)—a formulation that enables physically plausible yet flexible modeling of tool dynamics. Additionally, the paper introduces Pose-Informed Point Fusion (PIPF), a novel Gaussian initialization strategy that leverages multi-frame depth fusion using 6-DoF tool poses, addressing a key bottleneck in GS-based methods for handling fast-moving, non-rigid objects.
By integrating these components into a unified 4D Gaussian Splatting framework, the method achieves state-of-the-art reconstruction accuracy and real-time rendering performance on two public datasets. This makes T²GS a technically innovative and practically promising solution for intraoperative scene understanding in minimally invasive surgery.
- 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 goes beyond the typical use of Gaussian Splatting (GS) by structurally extending and re-designing the GS pipeline to tackle the unique challenges of dynamic surgical scene reconstruction. Rather than simply applying GS to tissue deformation, the authors introduce a dedicated motion modeling framework (ETMM) that decomposes surgical tool motion into global rigid-body trajectories and local non-rigid shape changes—an original and well-motivated formulation.
The proposed Pose-Informed Point Fusion (PIPF) module is another novel contribution, enabling robust initialization of tool-specific Gaussians by leveraging multi-frame depth fusion based on estimated 6-DoF poses. This directly addresses a key limitation of standard GS methods, which struggle to handle moving instruments with standard SfM-based initialization.
Compared to prior GS-based methods that use a single deformation field to represent the entire scene, this work introduces a multi-object, multi-scale motion modeling structure, making it highly suitable for complex MIS environments with interacting tissues and instruments.
- 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 paper demonstrates promising real-time performance and high reconstruction quality, suggesting potential for intraoperative applications. However, it lacks direct evaluation in actual clinical workflows. Important clinical use cases—such as surgical navigation, operator assistance, or skill assessment—are not explored or validated. Additionally, there is no user study or feedback from clinicians, making the practical utility of the method in real surgical settings uncertain. Further validation in collaboration with medical professionals would strengthen the clinical relevance of the work.
- 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.
- 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.
(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 technically novel and well-engineered extension of Gaussian Splatting for dynamic surgical scenes. The proposed ETMM and PIPF modules effectively address tool motion modeling and initialization challenges, demonstrating clear advantages over prior work. Strong experimental results, real-time performance, and well-organized presentation support its relevance and quality, making it a candidate for 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.
The authors provide a well-structured and respectful rebuttal that effectively addresses the major concerns raised by the reviewers. They clarify misunderstandings around experimental results (e.g., Tab. 1 values and tool-only reconstruction), offer justifications for the design choices (such as the focus on tool modeling rather than tissue-only baselines), and acknowledge limitations where appropriate (e.g., lack of clinical validation and 6DoF pose evaluation).
Importantly, the authors demonstrate a strong understanding of the reviewers’ comments and respond with clarity and transparency, without overpromising or deflecting criticism. Their revisions to figures, clarifications in the manuscript, and plans for future clinical and quantitative validation add credibility and reflect thoughtful engagement with the feedback.
Author Feedback
We thank the reviewers for the constructive feedback and appreciate the consensus that our method is well-motivated, innovative, and effective, particularly in 4D reconstruction quality and efficiency. We confirm our plan to release the code upon acceptance.
Major Concerns:
Misunderstanding of Experiment Values(R1-7.1, R3-7.2): Sincerely thank the reviewers for the concerns raised by the bracketed values in Tab.1, and apologize for the confusion caused by the table layout due to our improper pursuit of compactness. As noted in Tab.1’s caption, these bracketed values per row represent PSNR-before/PSNR-after and SSIM-before/SSIM-after. Specifically, “before” refers to an extra experiment of Deform3DGS(which we adopt for tissue reconstruction in TTGS) on the tissue-only task, namely only reconstructing tissue by masking tools; while “after” corresponds to our TTGS on the whole scene reconstruction task (reconstructing both tissue and tool). These bracketed values are reported on tissue-area only, with an intention to show that TTGS’s overall improvements do not substantially impair tissue reconstruction, as briefly discussed by “without impairing…” in Sec.3.2. We have revised the table layout, specified the notations, and added the intention for a clear presentation. Also, sorry for the unclear statement, which led to R1’s misunderstanding on missing metrics for TTGS on moving tool reconstruction. The reported results ETMM(ours) in Tab.1 are exactly the tool-only reconstruction results using our method(TTGS). We revised our manuscript and made it clearer.
- Suggested Experiments:
- Cross-methods Eval. on tissue recon. task(R3-7.1): Thanks for the valuable concern on the tissue area. For Deform3DGS, we conducted such an experiment(metrics indicated as “before” in Tab.1) to show our TTGS’s overall improvements do not substantially impair tissue recon., as we’ve clarified in the above major concerns Point 1. For the other baselines, we intentionally omitted further eval. as the contributions of our work do not lie in the chosen tissue recon. method (we adopted unchanged from Deform3DGS); therefore, across-methods comparison on the tissue-only task is less prioritized.
- 6DoF Tool Pose Evaluation(R3-7.3): Thanks for the valuable suggestion on eval. of the intermediate optimized 6DoF poses. We indeed observed visually improved tool poses, but opted to omit them and highlighted our main contributions in scene reconstruction quality due to the page limit. Recognizing our method’s strong potential for accurate surgical tool pose estimation, we are collecting data with GT tool pose for future work focused on this aspect with a comprehensive quantitative evaluation.
- Clinical Validation(R2-7): We thank the reviewer for the constructive insights and emphasis on further clinical validation. We plan to employ our method for skill assessment through collaboration with medical professionals, alongside broader downstream tasks.
- Method Figure Quality(R1-7.2): Sincerely thank the reviewer for pointing out the under-quality of Fig.1, and we deeply regret that Fig. 1 is not sufficiently helpful in understanding our approach. We carefully improved Fig. 1 with detailed and consistent notations (including correcting ‘L_color’ and ’L_depth’ to ‘L_C’ and ‘L_D’) of all present elements, a better layout, and more detailed captions. Hope the improvements to Fig. 1 can also clarify the intended message in Fig. 1(b) — it should illustrate our tool motion model presented in Sec. 2.4—Temporal tool motions are separated into global motion and local shape change, modeled separately by pose-guided global trajectory modeling(P-GTM) and local shape-change modeling(LSM), together forming our tool motion module(ETMM).
Minor Concerns: Typo and incomplete sentence(R1, R3): Thank the reviewers for the kind reminder on typos and incomplete sentences. We sincerely apologize for these oversights, have carefully corrected them all with a thorough internal review.
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’
This paper proposes a novel framework, T²GS, for dynamic surgical scene reconstruction by structurally extending the Gaussian Splatting (GS) pipeline to model both deformable tissues and dynamically moving surgical tools. The main technical contributions lie in the Efficient Tool Motion Modeling (ETMM) module, which separates global and local motion components of tool dynamics, and the Pose-Informed Point Fusion (PIPF) method for more robust Gaussian initialization. Together, these modules aim to address limitations of prior GS-based methods in representing fast-moving, non-rigid objects within surgical environments. Experimental results are presented on two public datasets, and the authors commit to code release upon acceptance.
Overall, the paper demonstrates a strong technical foundation, with a method tailored to the specific challenges of complex surgical environments. Reviewer #2 provided a firm accept and commended the originality and clarity of the work, recognizing the modular motion modeling as a substantial improvement over prior approaches. Reviewer #3 also leans positive, appreciating the methodological innovation and suggesting only incremental revisions and clarifications. These include additional tool pose evaluation and more structured comparisons on tissue-only metrics. The authors respond to these points in a measured and detailed rebuttal, clarifying Table 1, acknowledging the value of pose evaluation (with plans for future validation), and making improvements to Figure 1 and other presentation issues.
Reviewer #1 raises more concerns, pointing out formatting errors, unclear figure annotations, and a need for more careful experimental reporting. The reviewer’s score (3) is driven largely by these issues, which while valid, appear fixable and do not undermine the technical validity of the work. The rebuttal satisfactorily clarifies the metrics in Table 1, acknowledges oversights, and commits to revisions. Importantly, the reviewer does not question the core contributions or methodological soundness of the approach.
While the paper does not yet include clinical validation or tool pose ground truth evaluation, it sets the groundwork for these directions and explicitly articulates them as future work. The reconstruction results on tool dynamics—a central contribution—are quantitatively and qualitatively strong. The novelty and specificity of the motion modeling justify acceptance despite these limitations, especially in the context of computer-assisted interventions where robust modeling of surgical tools remains a key challenge.
Given the well-structured method, promising results, and thoughtful rebuttal, this AC recommend 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A