ColonSplat: Reconstruction of Peristaltic Motion in Colonoscopy with Dynamic Gaussian Splatting

Weronika Smolak-Dyżewska; Joanna Kaleta; Diego Dall'Alba; Przemysław Spurek

cs.CVMar 6, 2026

ColonSplat: Reconstruction of Peristaltic Motion in Colonoscopy with Dynamic Gaussian Splatting

Authors: Weronika Smolak-Dyżewska, Joanna Kaleta, Diego Dall'Alba, Przemysław Spurek

Organizations: Jagiellonian University, Kraków, Poland · Warsaw University of Technology, Poland · Sano Centre for Computational Medicine, Kraków, Poland · University of Verona, Italy ·

Abstract

Accurate 3D reconstruction of colonoscopy data, accounting for complex peristaltic movements, is crucial for advanced surgical navigation and retrospective diagnostics. While recent novel view synthesis and 3D reconstruction methods have demonstrated remarkable success in general endoscopic scenarios, they struggle in the highly constrained environment of the colon. Due to the limited field of view of a camera moving through an actively deforming tubular structure, existing endoscopic methods reconstruct the colon appearance only for initial camera trajectory. However, the underlying anatomy remains largely static; instead of updating Gaussians' spatial coordinates (xyz), these methods encode deformation through either rotation, scale or opacity adjustments. In this paper, we first present a benchmark analysis of state-of-the-art dynamic endoscopic methods for realistic colonoscopic scenes, showing that they fail to model true anatomical motion. To enable rigorous evaluation of global reconstruction quality, we introduce DynamicColon, a synthetic dataset with ground-truth point clouds at every timestep. Building on these insights, we propose ColonSplat, a dynamic Gaussian Splatting framework that captures peristaltic-like motion while preserving global geometric consistency, achieving superior geometric fidelity on C3VDv2 and DynamicColon datasets. Project page: https://wmito.github.io/ColonSplat

Figures & tables

Figure 1: The C3VDv2 dataset contains realistic colonoscopy sequences with substantial non-rigid deformations over time. The top two rows show captures from the endoscopic camera and the corresponding renders produced by different methods. The bottom row illustrates the global structure of the reconstructed colon. Peristaltic-like dynamic challenge baseline approaches; however, ColonSplat uniquely maintains a physically plausible global structure across timesteps. Please zoom in for details, see more examples in supplementary videos.

Figure 2: ColonSplat reconstructs dynamic 3D anatomy from colonoscopy video using estimated depth. A deformation model updates canonical Gaussians parameters at each time step for consistent dynamic reconstruction. Training uses RGB and depth supervision, with KNN and color regularization, to ensure accurate, artifact-free results.

	C3VDv2			DynamicColon
Method	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	CH $\downarrow$	HD95 $\downarrow$	MSE D $\downarrow$
EndoPlanar	27.229	0.738	0.499	28.195	0.902	0.332	0.497	1.055	0.2999
SurgicalGS	27.624	0.766	0.350	31.635	0.942	0.075	0.384	1.136	0.0079
Deform3DGS	27.419	0.745	0.395	31.635	0.942	0.109	0.383	1.143	0.0077
Endo4DGS	27.491	0.728	0.582	30.548	0.945	0.155	0.372	1.017	0.0080
Endo4DGX	24.525	0.707	0.521	29.722	0.921	0.322	0.433	1.133	0.0169

Table 1: Quantitative comparison and ablation study on C3VDv2 and DynamicColon. Compared to baselines ColonSplat achieve superior geometry fidelity while offering strong reconstruction results. Ablation is shown in three bottom rows. *Lack of

\Delta c

modeling caused major Gaussian drift and degraded CH and HD95 in one scene.

Figure 3: Qualitative comparison for C3VDv2 dataset. Please zoom in for details. ColonSplat achieves superior reconstruction quality compared to the baselines.

Figure 4: DynamicColon. The top row shows renders from test trajectory. The bottom row presents views from cameras positioned outside the colon. Competing methods produce significant artifacts that mimic deformations. ColonSplat accurately captures the 3D structure of the colon without such artifacts, even when viewed from outside.

Figure 5: Ablation study on C3VDv2 scene. Our proposed components significantly enhance realistic dynamic colon deformation. KNN regularization provides smooth tissue-like Gaussian surface across timesteps while our proposed constraints eliminate geometric artifacts. Please zoom in for details.

Dynamic Gaussian Splatting

See all 14 papers

C3VD-DEFCOL: A Deformable Colonoscopy Dataset with Time-Resolved 3D Ground Truth and Realistic Appearance
Jun 5, 2026Ethan Luk, Mayank V. Golhar, Anthony Song +5Endoscopy 3D Reconstruction
Department of Biomedical Engineering, Johns Hopkins University, Baltimore, MD, USA · Instituto de Investigación en Ingeniería de Aragón (I3A), Universidad de Zaragoza, Zaragoza, Spain · Department of Computer Science, Johns Hopkins University, Baltimore, MD, USA
Colon3R: Cross-Domain 3D Reconstruction from Monocular Colonoscopic Video
Sep 21, 2026Zhihao Xing, Yingyu Wang, Liang Zhao +1Domain Adaptation 3D Reconstruction
Lesion-centered 3D mapping of colonoscopy procedures: validation of a hierarchical ensemble pipeline on public benchmark videos
Sep 15, 2026Hyunjun Kim, Hyeonwoo Na, Jaewoo LeeEndoscopy 3D Reconstruction
School of Computing, KAIST, Daejeon, Republic of Korea · Clinical Imaging Research Institute, Seoul, Republic of Korea · Division of Mechanical and Space Engineering, Faculty of Engineering, Hokkaido University, Sapporo, Japan +1

Jun 5, 2026J/K move · Enter open · S save

C3VD-DEFCOL: A Deformable Colonoscopy Dataset with Time-Resolved 3D Ground Truth and Realistic Appearance

Ethan Luk, Mayank V. Golhar, Anthony Song, Raúl Iranzo+4

Department of Biomedical Engineering, Johns Hopkins University, Baltimore, MD, USA · Instituto de Investigación en Ingeniería de Aragón (I3A), Universidad de Zaragoza, Zaragoza, Spain · Department of Computer Science, Johns Hopkins University, Baltimore, MD, USA

Endoscopy 3D Reconstruction Dynamic Scene Reconstruction Deformable Shape Reconstruction Endoscopic 3D Reconstruction Medical Image Benchmarks

Open paper AskPDF arXiv

3D reconstruction could improve colonoscopy by estimating mucosal coverage and alerting clinicians to missed regions during screening. However, algorithm development is limited as no current datasets provide both a realistic in vivo appearance and dense, time-resolved 3D ground truth, especially under non-rigid deformation. We present C3VD-DEFCOL, a framework and dataset for evaluating deformable colonoscopy reconstruction with paired geometry and realistic texture. Starting from C3VD/C3VDv2 colon meshes and camera trajectories, we generate controlled deformations of the colon surface, including peristaltic waves and centerline motion, and render per-frame depth, surface normals, optical flow, camera poses, and time-stamped 3D meshes. We then use the rendered geometry, primarily depth, to condition an LTX-2.3-based sim-to-real translation model that produces RGB clips with in vivo-like mucosal color, texture, vasculature, and specular appearance while preserving the underlying 3D scene structure. The resulting dataset contains 110 videos from 11 unique colon mesh geometries, with varying camera trajectories, appearances, and parameterized deformation regimes, including three peristaltic severity levels that serve as controlled evaluation axes. We evaluate the generated videos using appearance realism, geometric consistency, and temporal consistency metrics, and use the paired ground truth to benchmark the downstream task of pose estimation in deformable 3D reconstruction. Our experiments show how pose estimation error increases with increasing deformation severity, providing a controlled stress test that is not possible with existing in vivo datasets. Overall, C3VD-DEFCOL is designed as a reproducible, quantitative evaluation platform for testing deformable 3D reconstruction algorithms, with the goal of reducing the domain gap between synthetic datasets and in vivo colonoscopy.

Abstract

Dynamic Gaussian Splatting

Explore similar work

C3VD-DEFCOL: A Deformable Colonoscopy Dataset with Time-Resolved 3D Ground Truth and Realistic Appearance

Colon3R: Cross-Domain 3D Reconstruction from Monocular Colonoscopic Video

Lesion-centered 3D mapping of colonoscopy procedures: validation of a hierarchical ensemble pipeline on public benchmark videos