List of Papers Browse by Subject Areas Author List
Abstract
The visualization of volumetric medical data is crucial for enhancing diagnostic accuracy and improving surgical planning and education. Cinematic rendering techniques significantly enrich this process by providing high-quality visualizations that convey intricate anatomical details, thereby facilitating better understanding and decision-making in medical contexts. However, the high computing cost and low rendering speed limit the requirement of interactive visualization in practical applications. In this paper, we introduce ClipGS, an innovative Gaussian splatting framework with the clipping plane supported, for interactive cinematic visualization of volumetric medical data.
To address the challenges posed by dynamic interactions, we propose a learnable truncation scheme that automatically adjusts the visibility of Gaussian primitives in response to the clipping plane. Besides, we also design an adaptive adjustment model to dynamically adjust the deformation of Gaussians and refine the rendering performance. We validate our method on diverse medical data (including CT and anatomical slice data), and reach an average 36.635 PSNR rendering quality with 156 FPS and 16.1MB model size, showing the superior performance in rendering quality and efficiency.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/0380_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2025/supp/0380_supp.zip
Link to the Code Repository
https://github.com/med-air/ClipGS
Link to the Dataset(s)
N/A
BibTex
@InProceedings{LiChe_ClipGS_MICCAI2025,
author = { Li, Chengkun and Tong, Yuqi and Chen, Kai and Yang, Zhenya and Li, Ruiyang and Qiu, Shi and Chan, Jason Ying-Kuen and Heng, Pheng-Ann and Dou, Qi},
title = { { ClipGS: Clippable Gaussian Splatting for Interactive Cinematic Visualization of Volumetric Medical Data } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15969},
month = {September},
page = {126 -- 136}
}
Reviews
Review #1
- Please describe the contribution of the paper
The main contribution of this paper is the introduction of ClipGS, an interactive Gaussian Splatting framework designed specifically for cinematic visualization of volumetric medical data. The authors propose a Learnable Truncation scheme to dynamically control Gaussian primitive visibility relative to a clipping plane, significantly improving rendering accuracy and reducing artifacts. Additionally, they present an Adaptive Adjustment Model to dynamically deform Gaussian primitives, ensuring continuous visualization when interacting with internal anatomical structures.
- 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.
- better performance among other methods on a novel task
- a LT scheme that suits the interactive cinematic visualization task
- well prepared manuscript with detailed comparisions
- 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.
- Although I am not particularly interested in applying top-tier CV/CG techniques to the medical field for publications, this paper appears to involve 4DGS in the medical domain. Therefore, the authors should compare their work with existing 4DGS methods in the medical field, such as X2-Gaussian, TOGS, and others.
- The performance improvement is not substantial. I believe that a novel 4DGS-like method, such as MOSCA ST-4DGS or MODGS, could potentially achieve better performance on this task. If that is the case, the designed modules do not seem particularly inspiring.
- 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.
(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 purpose of this task is commendable; however, the novelty and actual significance of the proposed modules should be reconsidered. When PSNR exceeds 35, the visual difference becomes less noticeable.
- Reviewer confidence
Very confident (4)
- [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
The paper under review proposes a novel gaussian splat based method for displaying volumetric medical image data. An interactive cinematic visualization is implemented based on Gaussian splats that has support for adding clip planes.
- 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.
-gaussian splat adapted for medical image visualization -learnable truncation for improving the performance of the clipping -improvements to render performance via AAM model, rendering rates well above real time -large database of images based on the visible human project
- 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.
-how well is the method suited for rendering deformable structures? -an ablation that shows the impact of the initial number of views on the rendering quality should be added. Now the authors use 1000 images for training. What is the impact of this number on the image quality?
- 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?
Very well written paper that adapts Gaussian splats to be usable in the context if medical visualization. Several technical contributions are added on top of the standard Gaussian splat. These include improved rendering speed as well as consistent handling of clip planes. One can use a very expensive rendering algorithm as a start and train a splat that can then generate novel views and ensure real-time rendering times. This can be useful for surgical planning or anatomy teaching.
- Reviewer confidence
Very confident (4)
- [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
This paper presents a method for clipping 3D Gaussians for a given clipping plane. The optimization strategy includes a trainable clipping function using the straight-through estimator and continuous clipping using MLP with positional encoding. The method is compared with 4D Gaussian approaches and reports superior performance. The ablation study shows that the design choices are reasonable.
- 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-written and easy to follow. Clipping for real-time 3DGS for volumetric data is an emerging technology and a very interesting domain-specific application of 3DGS. The proposed training strategy provides flexibility to the rendering, and the design choices are technically sound and valid.
- 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 major weakness (if I needed to point out) is the fixed clipping plane direction. How flexibly the volume can be clipped or sliced is not clearly explained or demonstrated.
- 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.
(6) Strong Accept — must be accepted due to excellence
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The proposed approach and design choices for the training strategy are solid and well-validated. Adding some discussions on limitations would further strengthen the paper.
- Reviewer confidence
Very confident (4)
- [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
Author Feedback
We sincerely thank the AC and reviewers for their time and valuable feedback. We are encouraged by the overall positive reception. Reviewers described our method as “well-motivated,” “ well-written and easy to follow,” with “ superior performance”. We address the raised concerns point-by-point below:
Q: R1 wants to know the suitability for the deformable structure. A: The interactive visualization of deformable structure is a 5D problem compared to our current setting. Nevertheless, our framework is modular. Changing the backbone of our method from 3DGS to N-DG could help solve this problem. It will be a main research problem in our future work.
Q: R1 suggested showing the impact of training view’s number.
A: Thanks for the suggestion. In general, increasing the number of training views improves rendering quality but also increases preprocessing time. This presents a trade-off between preprocessing cost and visual fidelity. Limited by the paper length, we mainly focus on rendering performance under the fixed views, like other novel view synthesis works (e.g. 4DGaussians).Q: R2 mentioned the limitation of fixed clipping plane direction. A: We appreciate the observation. From the clipping plane along a fix direction, we actually could know every pixel value in the volume. A direct way to show random clipping surface is to query every pixels’ value. It is our planned work to integrate a random clipping plan rendering.
Q: R2 suggested more discussion on limitations. A: Thanks for the suggestion. Limited by paper space constraints in the current submission, we will add more discussion in the final version.
Q: R3 suggested to compare with more 4DGS-like methods like MOSCA, ST-4DGS & MODGS. A: To ensure a fair and representative evaluation, we selected a diverse set of baselines, including: 4DGaussians (dynamic parameter based), N-DG (high-dimensional latent representation), GauFRe (hybrid dynamic-static representation), and HexPlane (4DNeRF-style). These baselines capture a broad spectrum of 4DGS techniques, allowing us to effectively highlight the strengths of our proposed approach. As for the proposed reference methods, (1) The referenced methods are all dynamic-parameter-based methods, which are the variants of 4DGaussian. The absence of these methods does not affect our experiments and conclusions (2) These approaches rely on optical flow prior (e.g., via RAFT) of inter-frames along the temporal axis. In contrast, our setting introduces changes in visibility due to spatial clipping operations, which are not temporally continuous and therefore hard to introduce optical flow information. This makes direct application of such methods unsuitable for our task.
Q: R3 suggested to compare with X2-Gaussian & TOGS. A: Thank you for the suggestion. (1) X2-Gaussian was released on arXiv on March 27, 2025, which is after our submission deadline. (2) TOGS addresses a different problem setting. It focuses on modeling opacity variations in a temporally dynamic context, whereas our work is centered on the spatial visibility of Gaussian primitives under interactive clipping operations. Besides, TOGS actually is a type of variant of 4DGaussian (dynamic parameter based), which has been included in our experiments.
Q: To R3’s concerns about PSNR value. A: There are two reasons here for the high PSNR value in results: (1) We compute unmasked PSNR over the entire image, including background regions rendered as black, which causes the PSNR metric inflating. (2) Our method demonstrates significant visual improvement, particularly around the clipping surface, as illustrated in Fig. 2. However, the clipped region occupies only a portion of the frame, which also contributes to higher overall PSNR values. This concern is well advised, in the future work we shall consider developing better PSNR variant metrics to overcome the existing issues.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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