ZODIAC: Zero-shot Octree-based Diffusion for Anatomical Completion

Miruna-Alexandra Gafencu; Vlad Bratulescu; Yordanka Velikova; Mohammad Farid Azampour; Nassir Navab

cs.CVAug 25, 2026

ZODIAC: Zero-shot Octree-based Diffusion for Anatomical Completion

Authors: Miruna-Alexandra Gafencu, Vlad Bratulescu, Yordanka Velikova, Mohammad Farid Azampour, Nassir Navab

Organizations: Chair for Computer Aided Medical Procedures (CAMP) Technical University of Munich, Boltzmannstr. 3, 85748 Garching, Germany · Munich Center for Machine Learning (MCML), Munich, Germany · Konrad Zuse School of Excellence in Reliable AI (relAI), Germany

Abstract

Recovering the full 3D spine anatomy from intraoperative ultrasound is an ill-posed inverse problem, as the complete structure must be inferred from incomplete and noisy observations. Acoustic occlusions and limited field of view create large unobserved regions, while view-dependent artifacts lead to variability in expert annotations of the visible anatomy. Current supervised ultrasound shape completion methods rely on synthetically generated incomplete-complete paired data to learn conditional mappings under a predefined distribution of simulated occlusions. However, real intraoperative occlusions do not necessarily follow this distribution, which can limit generalization to patient data. As a result, accurate and robust completion from noisy partial observations remains an unsolved problem. We propose a zero-shot shape completion framework that reconstructs the entire lumbar spine from partial ultrasound observations without relying on simulated training data. To accommodate unseen and irregular patterns of missing structures, we introduce blended completion, a mechanism that integrates the learned anatomical prior with incoming partial geometry at inference time. The method learns a generative diffusion prior over full anatomical shapes represented in an adaptive octree structure, enabling efficient modeling of the complete spine in a single forward pass. Validation on phantom and volunteer data shows that decoupling completion from a predefined corruption distribution improves generalisation under real occlusions, outperforming a fully supervised variant by 22% on HD95 completion error. Code and data are available at https://github.com/miruna20/ZODIAC.

Figures & tables

Figure 1: Training pipeline. In Phase I (Representation Learning), a VAE encoder maps a geometric octree to a latent octree, which is reconstructed into a continuous SDF. In Phase II (Shape Generation), a two-stage denoising diffusion probabilistic model (DDPM) generates shapes from noise: a low-resolution UNet first denoises octree splitting signals to produce a geometric octree, followed by a high-resolution UNet that denoises the corresponding latent features. Finally, the resulting latent octree is decoded into a final SDF.

Figure 2: Zero shot shape completion mechanism. Within the LR diffusion stage, the partial input is re-noised and blended back at each denoising step, constraining the generation to remain consistent with the observed structure. The resulting geometric octree is then passed to the HR diffusion stage to denoise the latent features. Finally, the frozen VAE decoder produces the complete shape SDF.

Figure 3: Qualitative analysis of completed meshes. ZODIAC produces anatomically and geometrically coherent completions compared to SITD (baseline) and TP-ODIAC (fully supervised), particularly on volunteer acquisitions, where baselines yield fragmented or collapsed surfaces.

Method	Eval.	Phantom			Balgrist (all)			Balgrist (difficult)
Method	Eval.	CD $\downarrow$	F1 $\uparrow$	HD95 $\downarrow$	CD $\downarrow$	F1 $\uparrow$	HD95 $\downarrow$	CD $\downarrow$	F1 $\uparrow$	HD95 $\downarrow$
SITD (baseline, supervised)	WS	$5.38$ $\pm 0.86$	$0.362$ $\pm 0.029$	$7.53$ $\pm 0.75$	$14.94$ $\pm 3.35$	$0.257$ $\pm 0.018$	$16.11$ $\pm 1.42$	$19.60$ –	$0.242$ –	$18.00$ –
SITD (baseline, supervised)	VW	$15.77$ $\pm 4.00$	$0.193$ $\pm 0.029$	$8.25$ $\pm 1.21$	$52.17$ $\pm 28.42$	$0.119$ $\pm 0.027$	$13.96$ $\pm 4.54$	$69.02$ $\pm 29.87$	$0.119$ $\pm 0.017$	$15.90$ $\pm 4.30$
TP-ODIAC (ours, supervised)	WS	$4.23$ $\pm 1.61$	$0.411$ $\pm 0.103$	$6.11$ $\pm 1.24$	$5.49$ $\pm 2.46$	$0.434$ $\pm 0.060$	$7.78$ $\pm 3.37$	$7.35$ $\pm 2.15$	$0.398$ $\pm 0.051$	$10.26$ $\pm 3.19$
TP-ODIAC (ours, supervised)	VW	$9.69$ $\pm 4.93$	$0.240$ $\pm 0.073$	$5.70$ $\pm 1.66$	$14.78$ $\pm 13.92$	$0.231$ $\pm 0.047$	$7.02$ $\pm 3.92$	$19.93$ $\pm 17.49$	$0.211$ $\pm 0.040$	$8.51$ $\pm 4.57$
ZODIAC (ours, unsupervised)	WS	$5.22$ $\pm 0.98$	$0.368$ $\pm 0.038$	$7.37$ $\pm 0.70$	$5.64$ $\pm 0.47$	$0.371$ $\pm 0.023$	$7.74$ $\pm 0.45$	$5.99$ $\pm 0.44$	$0.353$ $\pm 0.020$	$8.02$ $\pm 0.48$

Table 1: Quantitative evaluation on Phantom (n=2) and Balgrist volunteer data. Balgrist (all) includes 6 subjects for our methods and 3 for SITD; Balgrist (difficult) is a subset of 3 subjects (URS36_D2, URS45_H2, URS45_R2), and SITD reconstructs only one of these. CD scaled by

10^{4}

(normalised units); F1 is the F-score at the evaluation threshold (higher is better); HD95 in mm. All values for our methods are reported as mean

\pm

standard deviation, where each subject’s metric is first averaged over 10 stochastic samples and the mean and standard deviation are then taken across subjects. Eval.: WS = Whole Spine, VW = Vertebra-wise.

3D Shape Completion

3D Shape Completion reconstructs missing geometry from partial scans, images, or sensor measurements for perception, modeling, and medical analysis.

See all 23 papers

Mar 9, 2026

OSCAR: Occupancy-based Shape Completion via Acoustic Neural Implicit Representations

Accurate 3D reconstruction of vertebral anatomy from ultrasound is important for guiding minimally invasive spine interventions, but it remains challenging due to acoustic shadowing and view-dependent signal variations. We propose an occupancy-based shape completion method that reconstructs complete 3D anatomical geometry from partial ultrasound observations. Crucially for intra-operative applications, our approach extracts the anatomical surface directly from the image, avoiding the need for anatomical labels during inference. This label-free completion relies on a coupled latent space representing both the image appearance and the underlying anatomical shape. By leveraging a Neural Implicit Representation (NIR) that jointly models both spatial occupancy and acoustic interactions, the method uses acoustic parameters to become implicitly aware of the unseen regions without explicit shadowing labels through tracking acoustic signal transmission. We show that this method outperforms state-of-the-art shape completion for B-mode ultrasound by 80% in HD95 score. We validate our approach both in-silico and on phantom US images with registered mesh models from CT labels, demonstrating accurate reconstruction of occluded anatomy and robust generalization across diverse imaging conditions. Code and data will be released on publication.

Magdalena Wysocki, Kadir Burak Buldu, Miruna-Alexandra Gafencu +2

Chair for Computer Aided Medical Procedures (CAMP) Technical University of Munich, Boltzmannstr. 3, 85748 Garching, Germany · Munich Center for Machine Learning (MCML), Munich, Germany · Konrad Zuse School of Excellence in Reliable AI (relAI), Germany

Implicit Neural Representations 3D Reconstruction 3D Medical Imaging Ultrasound Imaging 3D Shape Completion

Ask PDF

Sep 10, 2026

UBone3D: Physics-Rectified Conditional Flow Matching for Anatomical 3D Shape Completion from Ultrasound

Three-dimensional ultrasound (US) is a safe, radiation-free complementary modality to CT and X-rays for longitudinal monitoring, yet its segmentation-derived partial point clouds are extremely artifact-laden. Consequently, it is challenging to recover a clean and complete anatomical structure from such US point clouds. In this paper, we present UBone3D, a novel framework based on physics-rectified conditional flow matching (CFM) that performs point cloud completion directly from partial US observations. UBone3D models deterministic physics artifacts (e.g., surface thickening, streaking, dropouts) via a simulated physics proxy, and introduces test-time physics rectification to steer the shape completion. At inference, the completion is jointly steered by two decoupled forces: (1) anatomical plausibility enforced by a CT-trained generative shape prior, BoneFM, and (2) physics consistency enforced by USimNet in the ultrasound formation space. Extensive experiments on simulated and in-vivo data demonstrate significant improvements in reconstruction accuracy and anatomical fidelity over existing baselines.

Weiying Chen, Yuchong Gao, Siyuan Li +3

University of Alberta, Edmonton, AB, Canada

Physics-Informed Generative Modeling Point Cloud Completion Conditional Flow Matching 3D Medical Imaging 3D Shape Completion Physics-Informed ML

Ask PDF

May 26, 2026

DinoComplete: 3D Shape Completion with Distilled Semantic Priors and State Space Models

3D shape completion from partial scans remains challenging for unseen categories and noisy real-world observations, where geometry alone is often insufficient for inferring missing structure. We present DinoComplete, a deterministic and efficient shape completion framework that augments geometric reconstruction with voxel-aligned semantic priors distilled from DINO features. First, we construct multi-view DINO feature volumes aligned with ShapeNet data and train a student network to predict dense semantic features directly from incomplete shapes. These predicted features capture global structure and part-aware semantic context while remaining aligned with the underlying geometry. We then integrate these distilled features into a completion network, where geometric and semantic voxel representations are fused through voxel state-space modeling. To enable efficient long-range reasoning without sacrificing resolution, we introduce a multi-scale voxel Mamba module that refines the fused features by combining full-grid and chunk-wise sequence modeling. Experiments on unseen ShapeNet categories and ScanNet objects show that DinoComplete achieves stronger completion quality than prior deterministic and generative based completion methods while using fewer parameters, requiring lower memory, and achieving faster inference. Our results demonstrate that distilling semantic priors from visual foundation models improves generalization and robustness in 3D shape completion.

Furkan Mert Algan, Eckehard Steinbach

Chair of Media Technology · Munich Institute of Robotics and Machine Intelligence · School of Computation Information and Technology, Technical University of Munich

3D Reconstruction 3D Shape Completion State Space Models

Ask PDF

Abstract

3D Shape Completion

Explore similar work

OSCAR: Occupancy-based Shape Completion via Acoustic Neural Implicit Representations

UBone3D: Physics-Rectified Conditional Flow Matching for Anatomical 3D Shape Completion from Ultrasound

DinoComplete: 3D Shape Completion with Distilled Semantic Priors and State Space Models