LatticeRank: Spacing-Ranked Consistency for Native-Grid 3D Medical Image Segmentation

Jingjia Yan; Yue Zhuo; Zhonghua Xie

cs.CVOct 20, 2026

LatticeRank: Spacing-Ranked Consistency for Native-Grid 3D Medical Image Segmentation

Authors: Jingjia Yan, Yue Zhuo, Zhonghua Xie

Organizations: Huizhou University, Huizhou 516007, Guangdong, China

Abstract

Variable voxel spacing changes more than image resolution: it alters the discrete sampling lattice, tensor geometry, and anatomical evidence available to a 3D segmentation network. While conventional pipelines restore images to a canonical grid, direct-grid deployment requires the model itself to tolerate these changes. We observe that candidate grids are not exchangeable views -- a coarse grid removes more structural information than a mildly resampled one -- yet standard consistency objectives assign them equal influence. We therefore propose LatticeRank, a training-only framework that aligns candidate posteriors in physical space and allocates consistency supervision by voxel-spacing order: among endpoint-matched candidates, the auxiliary loss concentrates on the coarsest requested grid, while inference retains a single network forward with no candidate branch or ensemble. On MSD Task04, LatticeRank improves coarse-grid Extra Dice over mean consistency by 0.03436 across five seeds while meeting the prespecified reference-grid noninferiority criterion. A controlled ladder shows that grid exposure provides the largest recovery, posterior consistency adds a second gain, and spacing-ranked allocation yields the strongest direct-grid result; requested spacing thus provides a simple, model-independent ordering signal for direct-grid consistency learning.

Figures & tables

Figure 1: LatticeRank training and deployment. Endpoint-matched candidate grids share the segmentation network, and their posteriors are aligned to R1 before consistency is measured against a detached reference posterior. Requested spacing orders the candidates from fine to coarse; the coarsest candidate supplies the auxiliary loss. Candidate branches and alignment operations are removed at inference, leaving one direct-grid forward.

Figure 2: Controlled ladder on 40 held-out Task04 cases and seeds 17/29/41. Candidate exposure supplies the largest direct-grid recovery, aligned posterior consistency adds a second gain, and LatticeRank (S) achieves the strongest coarse-grid result among direct methods. Restoration remains the accuracy-oriented path.

Method	R1	Extra	Worst	Grid-AUC	Train h
Single	0.88169	0.02102	0.00885	0.21639	0.63
Restore	0.88305	0.70226	0.63056	0.86056	2.80
DirectLabel	0.88348	0.25414	0.11299	0.60313	2.69
C	0.88365	0.33715	0.17547	0.65277	1.85
D	0.88417	0.35523	0.18959	0.66329	1.84
S (LatticeRank)	0.88344	0.37855	0.20926	0.68032	1.60

Table 1: Held-out ladder. Extra is mean mapped Dice at 4/5 mm; Worst is the minimum mapped Dice over 1.5–5 mm. Grid-AUC measures agreement with the R1 prediction. For Extra, Worst, and Grid-AUC, best values are bold and second-best values are underlined.

Figure 3: Five-seed S-versus-C evidence. LatticeRank improves Extra Dice over uniform consistency, while the R1 difference remains above the prespecified noninferiority margin. Seed-level effects show the variation across training runs, including the negative effect for seed 41.

Figure 4: Unmodified seed-17 qualitative cases. Rows show a 4-mm case, a 5-mm median-bracketing case, and the lowest 5-mm $S-\max(C,D)$ case. Columns show input, ground truth, restoration, C, D, and LatticeRank (S). Foreground Dice (restore/C/D/S) is 0.788/0.479/0.510/0.564, 0.511/0.160/0.174/0.210, and 0.520/0.178/0.192/0.175 by row. Blue and orange denote ground-truth and prediction contours.

Figure 5: Candidate-set sensitivity and same-dataset backbone comparison. Left: post-selection analyses of candidate count and training range on the three-seed held-out panel. Right: C/D/S with a Plain 3D U-Net on the development set. These supporting analyses do not redefine the main configuration.

Setting	Rule	R1	Extra	Worst	Grid-AUC
$K=2$ , 1–3 mm	S (frozen)	0.88344	0.37855	0.20926	0.68032
$K=3$ , 1–3 mm	S	0.88573	0.42039	0.25008	0.70160
$K=4$ , 1–3 mm	S	0.88361	0.41774	0.25150	0.70080
$K=2$ , 1–2 mm	S	0.88576	0.06705	0.02936	0.44836
Plain 3D U-Net	C	0.89631	0.41656	0.22196	0.72072
Plain 3D U-Net	D	0.89268	0.46165	0.27688	0.73891

Table 2: Candidate-set sensitivity and same-dataset backbone transfer. The upper block is post-selection held-out analysis; the lower block is development-only. Extra is mean mapped Dice at 4/5 mm.

Figure 6: Common-endpoint operator benchmark. Direct forwarding reduces warm-cache latency relative to restoration, but the accuracy gap grows on coarse grids and no peak-memory advantage is observed.

Medical Image Segmentation

Medical image segmentation delineates organs, lesions, and other structures in scans to support measurement, diagnosis, treatment planning, and research.

See all 786 papers

Towards Voxel Spacing Consistency for Medical Image Segmentation
Jun 30, 2026Xin You, Runze Yang, Minghui Zhang +63D Medical Image Segmentation
the Institute of Medical Robotics, Shanghai Jiao Tong University, 200240 Shanghai, China · CAMPAR, Technical University of Munich, 80333 Munich, Germany · the Department of Ultrasound, Shanghai Chest Hospital, 200052 Shanghai, China
Structure-Adaptive Sparse Diffusion in Voxel Space for 3D Medical Image Enhancement
Apr 20, 2026Hongxu Jiang, Fei Li, Boxiao Yu +43D Medical Imaging Medical Image Super-Resolution
Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL, USA · Department of Medicine, University of Florida, Gainesville, FL, USA · Department of Biomedical Engineering, University of Florida, Gainesville, FL +2
Topology-Driven Transferability Estimation for 3D Medical Vision Foundation Models
Jul 5, 2026Jiaqi Tang, Shaoyang Zhang, Fandong Zhang +3Medical Imaging Foundation Models 3D Medical Imaging
National Institute of Health Data Science, Peking University, Beijing, China · Institute of Medical Technology, Peking University, Beijing, China · State Key Laboratory of General Artificial Intelligence, Peking University, Beijing, China +2

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

Towards Voxel Spacing Consistency for Medical Image Segmentation

Xin You, Runze Yang, Minghui Zhang, Hanxiao Zhang+5

the Institute of Medical Robotics, Shanghai Jiao Tong University, 200240 Shanghai, China · CAMPAR, Technical University of Munich, 80333 Munich, Germany · the Department of Ultrasound, Shanghai Chest Hospital, 200052 Shanghai, China

3D Medical Image Segmentation

Open paper AskPDF arXiv

Volumetric medical image segmentation is essential for both preoperative diagnosis and intraoperative guidance. While recent years have witnessed rapid progress in segmentation architectures, comparatively little attention is paid to the physical voxel spacing of anatomical data. Indeed, volumetric image resampling is a ubiquitous preprocessing step before segmentation, yet its interaction with downstream segmentation has not been systematically exploited. In this work, we study the correlation between image resampling and segmentation, and propose Consispace, a semantic-aware resampling framework that achieves consistent voxel spacing in the axial direction while preserving anatomical and semantic consistency. Consispace introduces an ODE-based anatomical constraint to model inter-slice dynamics with a continuous interpolator, enabling faithful reconstruction under complex anatomical transitions beyond discrete interpolation. To further couple resampling with segmentation objectives, we leverage dense features from a pretrained vision model to build intra-slice semantic correlation maps and inject class-wise semantic consistency via feature reweighting during resampling. Both intra-slice and inter-slice constraints are integrated into an implicit neural network, supporting arbitrary-scale resampling. Extensive experiments on multiple datasets demonstrate that Consispace achieves superior reconstruction quality and perceptual fidelity, produces smoother inter-slice anatomy, and improves downstream segmentation performance when used as a preprocessing step.

Abstract

Medical Image Segmentation

Explore similar work

Towards Voxel Spacing Consistency for Medical Image Segmentation

Structure-Adaptive Sparse Diffusion in Voxel Space for 3D Medical Image Enhancement

Topology-Driven Transferability Estimation for 3D Medical Vision Foundation Models