AD-GS: Anatomical Density-guided 3D Gaussian Splatting for Sparse-View Dental CBCT Reconstruction

Zonglin Li; Kexian Xie; Feiran Li; Chu Zhou; Yixing Huang; Ruiyang Chen; Zhanyu Ma; Heng Guo

cs.CVOct 10, 2026

AD-GS: Anatomical Density-guided 3D Gaussian Splatting for Sparse-View Dental CBCT Reconstruction

Authors: Zonglin Li, Kexian Xie, Feiran Li, Chu Zhou, Yixing Huang, Ruiyang Chen, Zhanyu Ma, Heng Guo

Organizations: Beijing University of Posts and Telecommunications · Department of Stomatology, Peking University Third Hospital · Independent Researcher · National Institute of Informatics · Institute of Medical Technology, Peking University

Abstract

Sparse-view Cone-Beam Computed Tomography (CBCT) reconstruction is essential for reducing radiation dose and scan effort, but remains underconstrained due to limited projection information. Recent 3D Gaussian Splatting (3DGS)-based CBCT reconstruction methods rely on gradient-driven densification, which becomes unstable under sparse acquisition and often leads to over-smoothed structures. We propose AD-GS, an Anatomical Density-guided 3D Gaussian Splatting framework that formulates Gaussian densification as an attenuation-aware Markov Chain Monte Carlo (MCMC) sampling process. By interpreting Gaussians as samples from an underlying anatomical attenuation distribution, AD-GS adaptively concentrates representation capacity in high-attenuation regions, enabling sharper structural boundaries under sparse projections. To further constrain reconstruction, we incorporate a semantic regularization based on zero-shot features from Segment Anything in Medical Images, enforcing multi-view consistency through semantic priors. Experiments on the ToothFairy3 dental CBCT dataset demonstrate that AD-GS achieves superior structural fidelity and boundary preservation compared with existing 3DGS-based approaches under sparse-view settings within the dental and maxillofacial setting. Code is available at https://github.com/yulinjiaxue/AD-GS-BMVC2026.

Figures & tables

Figure 1: AD-GS achieves superior fidelity and sharper details than both the 3DGS-based method $R^{2}$ -Gaussian [ 22 ] and classical analytical method FDK [ 8 ] in sparse (30 views) CBCT reconstruction.

Figure 2: Overview of the AD-GS framework. Starting from FDK initialization, Radiative Gaussians are optimized via projection-consistency. The reconstruction is enhanced by anatomical density-guided MCMC sampling for dental structures and MedSAM-based semantic regularization for structural connectivity.

Figure 3: Visualization of MedSAM features via PCA. The compressed semantic maps delineate clear dental structures and boundaries, serving as a structural prior beyond raw intensities.

Figure 4: Comparison on CBCT reconstruction under different numbers of views. Ours are closest to the GT, especially near the region containing hard tissue.

Views	Metric	FDK [ 8 ]	SART [ 1 ]	ASD-POCS [ 17 ]	SAX-NeRF [ 5 ]	$R^{2}$ -Gaussian [ 22 ]	Ours
20	DSC $\uparrow$	80.39	76.73	78.61	25.25	78.75	88.26
	NSD $\uparrow$	83.17	79.23	82.09	30.27	82.17	93.69
	clDice $\uparrow$	79.94	76.98	79.50	28.13	78.48	90.05
	ASSD $\downarrow$	88.50	94.65	94.06	392.47	91.14	44.03
	PSNR $\uparrow$	17.67	27.67	27.44	27.19	28.81	30.78
30	DSC $\uparrow$	88.02	86.01	87.59	62.43	89.97	92.89

Table 1: Quantitative comparison of tooth reconstruction performance across different view counts. The best results are highlighted in bold , and the second-best results are underlined . DSC, NSD, and clDice are reported in (%), while ASSD is in (

10^{-2}

mm) and PSNR is in dB.

Figure 5: Evaluation on Downstream Task: Tooth Segmentation [ 19 ] . Side and frontal views of segmentation results obtained from different CBCT reconstructions. Geometric errors with respect to ground truth (GT), measured by SSD, are visualized using color maps. Our volume yields the most complete tooth segmentation and lowest SSD.

Views		20				30				50
Metric		DSC $\uparrow$	NSD $\uparrow$	clD $\uparrow$	ASSD $\downarrow$	DSC $\uparrow$	NSD $\uparrow$	clD $\uparrow$	ASSD $\downarrow$	DSC $\uparrow$	NSD $\uparrow$	clDice $\uparrow$	ASSD $\downarrow$
Exp 1: Robustness against Initialization (Tested on Noisy Data)
$R^{2}$ -Gaussian [ 22 ]	FDK	80.70	83.50	78.60	74.14	91.61	96.02	93.68	26.08	95.57	99.11	97.52	12.87
$R^{2}$ -Gaussian [ 22 ]	Random	76.27	77.73	74.98	98.29	90.83	95.08	92.79	29.10	95.36	99.08	97.45	13.24
Ours	FDK	90.53	95.30	92.00	28.75	94.13	98.83	96.42	16.89	95.74	99.56	97.79	12.45
Ours	Random	89.98	94.55	91.50	29.68	94.27	98.67	96.20	16.69	95.92	99.66	97.79	11.97

Table 2: Quantitative evaluation under different initializations, noise levels, and ablation settings.

Cat.	Method	20 Views	30 Views	50 Views
Trad.	FDK [ 8 ]	1.3	1.7	1.2
	SART [ 1 ]	77.5	115.1	195.0
	ASD-POCS [ 17 ]	79.7	122.0	198.0
NeRF-based	SAX-NeRF [ 5 ]	11,210	16,092	25,828
GS-based	$R^{2}$ -Gaussian [ 22 ]	854	855	833
	AD-GS (w/o SSR)	1,439	1,413	1,348

Table 3: Training time (seconds) on an NVIDIA RTX 4090. AD-GS adds manageable overhead over R 2 -Gaussian while remaining an order of magnitude faster than NeRF-based methods.

Explore similar work

Residual Gaussian Splatting for Ultra Sparse-View CBCT Reconstruction
Apr 30, 2026Jian Lin, Jiancheng Fang, Shaoyu Wang +43D Gaussian Splatting Cone-Beam CT
School of Information Engineering, Nanchang University, Nanchang 330031, China · Laboratory of Image Science and Technology, the School of Computer Science and Engineering, and the Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications, Ministry of Education, Southeast University, Nanjing 210096, China
X-Splat: Gaussian Splatting for 3D CBCT Generation from Single Panoramic Radiograph
Jul 2, 2026Tomasz Szczepański, Szymon Płotka, Michal K. Grzeszczyk +23D Gaussian Splatting Cone-Beam CT
Sano Centre for Computational Medicine, Poland · Jagiellonian University, Poland · Warsaw University of Technology, Poland +2
Shape-guided Gaussian Splatting for Sparse-View X-ray 3D Reconstruction
Sep 9, 2026Pranav Poudel, Florence Dell'Aniello Picard, Nairouz Shehata +23D Gaussian Splatting 3D Medical Imaging
Polytechnique Montréal, Canada · Mila - Quebec AI Institute, Canada · CHUM - University of Montreal Hospital, Canada

Apr 30, 2026J/K move · Enter open · S save

Residual Gaussian Splatting for Ultra Sparse-View CBCT Reconstruction

Jian Lin, Jiancheng Fang, Shaoyu Wang, Changan Lai+3

School of Information Engineering, Nanchang University, Nanchang 330031, China · Laboratory of Image Science and Technology, the School of Computer Science and Engineering, and the Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications, Ministry of Education, Southeast University, Nanjing 210096, China

3D Gaussian Splatting Cone-Beam CT Sparse-View CT Reconstruction

Open paper AskPDF arXiv

While 3D Gaussian splatting (3DGS) offers explicit and efficient scene representations for cone-beam computed tomography reconstruction, conventional photometric optimization inherently suffers from spectral bias under ultra sparse-view conditions, leading to over-smoothing and a loss of high-frequency anatomical details. Since wavelet transforms provide rich high-frequency information and have been widely utilized to enhance sparse reconstruction, this work integrates wavelet multi-resolution analysis with 3DGS. To circumvent the mathematical mismatch between the strict non-negativity of physical X-ray attenuation and the bipolar nature of high-frequency wavelet coefficients, we propose Residual Gaussian Splatting (RGS). Methodologically, we introduce a spectrally-decoupled Gaussian representation that stratifies the volumetric field into a geometric base component and a residual detail component. This decomposition systematically transforms explicit high-frequency fitting into a physically consistent, implicit residual compensation task. Furthermore, we devise a spectral-spatial collaborative optimization strategy to coordinate the interplay between geometric anchoring and texture refinement, effectively preventing spectral crosstalk. Extensive experiments on clinical datasets demonstrate that RGS enables the reconstructed images to capture highly refined geometric textures. It successfully resolves the trade-off between artifact suppression and detail preservation, yielding superior visual fidelity in complex trabecular and vascular structures compared to existing neural rendering baselines.

Abstract

Explore similar work

Residual Gaussian Splatting for Ultra Sparse-View CBCT Reconstruction

X-Splat: Gaussian Splatting for 3D CBCT Generation from Single Panoramic Radiograph

Shape-guided Gaussian Splatting for Sparse-View X-ray 3D Reconstruction