List of Papers Browse by Subject Areas Author List
Abstract
Reconstructing deformable 3D anatomy from endoscopic video is essential for surgical navigation and intraoperative guidance, yet remains challenging under the extreme illumination variations inherent to endoscopic imaging.
Co-axial near-field illumination, specular reflections on wet tissue, and automatic exposure changes violate the photometric consistency assumption underlying 3D/4D Gaussian Splatting (3DGS/4DGS), causing per-frame illumination variations to be entangled with Gaussian attributes, which degrades both geometry and rendered appearance.
Existing approaches either assume distant illumination incompatible with the endoscopic near-field setting, or apply learned appearance corrections without physically grounded decomposition.
We propose \textbf{LumiState}, which integrates Retinex-inspired reflectance illumination decomposition into 4D Gaussian Splatting for dynamic endoscopic reconstruction.
LumiState decomposes each frame into view-independent tissue reflectance and a per-frame illumination map at the Gaussian primitive level, ensuring that base Gaussians encode reflectance alone.
A two-stage training strategy and bright-channel prior further regularize the inherently ill-posed decomposition.
Experiments on the RLLS and EndoNeRF-EC benchmarks demonstrate state-of-the-art novel-view synthesis quality, with the largest gains under severe illumination variation, while enabling controllable illumination editing and temporally consistent reflectance rendering.
Code will be released upon acceptance.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2271_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/oDaiSuno/LumiState
Link to the Dataset(s)
N/A
BibTex
@InProceedings{SunDai_LumiState_MICCAI2026,
author = { Sun, Dai AND Guan, Huhao AND Zhang, Kun AND Liu, Yanzhe AND Xie, Xike AND Zhou, Shaohua Kevin},
title = { { LumiState: Retinex-inspired Illumination Decoupling in 4D Gaussian Splatting for Endoscopic Reconstruction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16892},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The main contribution of this paper is the integration of a Retinex-inspired reflectance–illumination decomposition into 4D Gaussian Splatting for dynamic endoscopic reconstruction. The method explicitly models each frame as the element-wise product of reflectance and illumination, and introduces an illumination estimation network together with Gaussian-level modulation to disentangle these components. A two-stage training strategy and additional regularization are used to stabilize the decomposition. The approach demonstrates consistent improvements over prior methods, particularly under strong illumination variations.
- 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 addresses the problem of photometric inconsistency caused by strong illumination variations in endoscopic reconstruction.
- The integration of a Retinex-inspired reflectance–illumination decomposition into 4D Gaussian Splatting is a simple and well-motivated idea.
- The per-Gaussian modulation of spherical harmonics (SH) coefficients conditioned on an illumination embedding is a novel design choice that enforces disentanglement directly within the 3D representation, rather than applying post-hoc 2D correction.
- The experimental results show consistent improvements over strong baselines, particularly under severe illumination variations.
- The ablation study is well conducted and provides useful insights into the contribution of each component.
- 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 Retinex formulation with view-independent R_t implicitly assumes a Lambertian model and may not adequately capture specular reflections, which are cited as a key motivation in the paper. While the additive term \delta_i is intended to account for such effects, no explicit analysis or validation is provided.
- Since ground-truth illumination is not available for real data, parts of the evaluation rely on qualitative results. Including validation on simulated data with known illumination would further strengthen the claims.
- While the method shows quantitative improvements over baselines, the performance gains are relatively modest.
- The approach is largely empirical, and the sensitivity to hyperparameters and the two-stage training schedule is not analyzed.
- The Retinex formulation with view-independent R_t implicitly assumes a Lambertian model and may not adequately capture specular reflections, which are cited as a key motivation in the paper. While the additive term \delta_i is intended to account for such effects, no explicit analysis or validation is provided.
- 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
- Duplicate references: [7] and [8] (3D Gaussian Splatting), [22] and [23] (LerPlane), [26] and [27] (EndoSurf) are the same entries and should be consolidated.
- Typo on Page 4 / Fig. 1: “Guassians” → “Gaussians. “
- 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 — marginally above the acceptance threshold, but would not mind if 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 addresses an important problem in endoscopic reconstruction under strong illumination variations, and the integration of a Retinex-based decomposition into 4D Gaussian Splatting is reasonable and well-motivated approach. The experimental results are solid, showing consistent improvements over baselines, and the ablation study is appropriately conducted. Although the method builds on a combination of existing ideas, the overall approach is simple and convincing. However, the evaluation of illumination decomposition is partly indirect, and the quantitative improvements are relatively modest.
- Reviewer confidence
Somewhat confident (2)
- [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 presents LumiState, a novel 4D Gaussian Splatting (4DGS) framework designed to address photometric inconsistency in dynamic endoscopic reconstruction. The core contribution is the integration of a Retinex-inspired reflectance-illumination decomposition. By employing a lightweight UNet for illumination estimation and a Fusion MLP for spatially adaptive SH coefficient modulation, the method successfully decouples intrinsic tissue reflectance from frame-varying near-field illumination. Experiments on RLLS and EndoNeRF-EC benchmarks demonstrate superior performance in synthesis quality and illumination control, while maintaining a real-time rendering speed of 84 FPS.
- 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 problem is important and well-motivated by realistic endoscopic scenarios. 2.Compared with pure affine appearance correction, the Retinex-inspired reflectance-illumination decomposition is more interpretable. 3.The experimental results including quantitative and qualitative evaluations show consistent improvements over exsiting methods. The ablation studies are reasonably comprehensive.
- 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 Retinex model $I = R \odot L$ typically assumes Lambertian surfaces. However, endoscopic scenes are characterized by intense specular reflections on wet tissues. Can authors provide more analysis on how well the Retinex decomposition holds in such non-Lambertian conditions, and whether the method can handle strong specular highlights effectively? 2.The paper uses 4D Gaussian Splatting as the underlying representation, and showing the the rendering results of consecutive frames will be more convincing. (Authors only introduce the fps results, but not the visual results of consecutive frames.) 3.The references contain apparent duplicates (e.g., repeated entries for 3DGS/LerPlane/EndoSurf), which weakens presentation quality.
- 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 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.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if 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 has a meaningful problem setup, a reasonable method, and positive empirical trends, especially render results under high illumination variance. Above all, I would like to accept this paper, but I encourage the authors to address the above weaknesses in the rebuttal and final version.
- 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 LumiState, a Retinex-inspired 4D Gaussian Splatting (4DGS) framework designed to address severe illumination variations in dynamic endoscopic scenes.
- 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.
LumiState explicitly incorporates Retinex theory, leading to stronger physical interpretability in the illumination-reflectance decomposition. The authors introduce several effective technical components, including modulation of SH coefficients via positional encoding to propagate illumination information across Gaussians, a two-stage training scheme, and the use of bright-channel prior to avoid degenerate solutions. Experimental validation on the RLLS and EndoNeRF-EC datasets shows that this method delivers industry-leading reconstruction quality under conditions of drastic changes in lighting.
- 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.In real endoscopic surgery, frequent insertion and movement of surgical instruments are among the primary causes of abrupt illumination changes. The proposed “temporal illumination completion” (Equation 5) relies on linear interpolation to recover illumination states. However, tool occlusion is typically sudden and highly non-linear. Simple linear interpolation may fail to handle instantaneous brightness jumps caused by tools entering or exiting the field of view. It would be valuable if the authors could demonstrate the rendered reflectance (Rt) quality on datasets containing frequent tool interactions to better assess the clinical practicality of the method. 2.In Equation (4), the additive compensation term δi is introduced to model subsurface scattering and other residuals. While helpful for rendering quality, its dependence on the illumination embedding “e_t” may introduce additional ambiguity between global illumination effects and material-dependent residuals. This could somewhat weaken the physical motivation behind the Retinex-inspired decomposition. The authors should provide an ablation study by removing δ_i and discuss its impact on both rendering quality and the cleanliness of the reflectance-illumination separation (for example, using metrics such as R-Var). A brief analysis in the limitations section would also strengthen the paper. 3.reference [7][8], [22][23], [26][27]are the same. 4.Figure 1 should be presented and clearly explained. 5.Efficiency of the proposed method better to be shown in a Table. 6.Is a better illumination correction result (evaluated by No-reference image quality metric in the paper ) leads to better reconstruction result?
- 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.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if 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 addresses an important issue: Reconstructing deformable 3D anatomy from endoscopic video under the extreme illumination variations inherent to endoscopic imaging. The paper propose a reasonable solution, which integrates Retinex-inspired reflectance illumination decomposition into 4D Gaussian Splatting. Experiments on two dataset show the effectiveness of the proposed method. However, more clearly presentation of the method, and more deep analysis of the proposed method and effectivenss of the proposed module is essential. And the manuscript contains some presentation typos.
- 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
Author Feedback
N/A
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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.
All reviewers are in favour of accepting this work, citing its interesting and novel combination of Retinex theory with Gaussian Splatting, as well as its strong experimental results and convincing performance. Due to unanimous agreement, I recommend that this submission be accepted, and that the authors consider fixing minor issues in the camera ready version.
