aiqbench analysis
This research introduces EAMS, a lightweight equivariant neural network framework designed to maintain robust anatomical mesh segmentation despite geometric perturbations like rotation. By leveraging Equivariant Mesh Neural Networks (EMNN) and anatomy-aware priors, the model achieves stability on irregular surface geometries where standard methods fail under test-time pose changes. The approach demonstrates that intrinsic mesh descriptors combined with global context can deliver competitive performance across diverse clinical applications without sacrificing efficiency.
Augmented Equivariant Mesh Networks for Anatomical Segmentation
arXiv:2605.08172v2 Announce Type: replace-cross Abstract: Anatomical mesh segmentation requires models that operate directly on irregular surface geometry while remaining robust to changes in coordinate pose across meshes of varying resolution. Existing task-specific mesh and point-cloud methods are not equivariant, and can degrade sharply under test-time perturbation, for example dropping by 28-30 IoU points on 3D-IOSSeg at $40^\circ$ rotation even with matched training augmentation. We present EAMS, an Equivariant Anatomical Mesh Segmentor built on Equivariant Mesh Neural Networks (EMNN), and evaluate it in four dataset settings across three clinical application areas, spanning edge-, vertex-, and face-level supervision. We combine intrinsic mesh descriptors with anatomy-aware priors, including PCA-derived frames for dental arches and liver surfaces, and augment message passing to provide lightweight global context. Across intracranial aneurysm and intraoral segmentation, EAMS variants are competitive with specialized baselines on unperturbed inputs while remaining stable under geometric perturbations, and on liver surfaces they expose a favorable trade-off between canonical-pose accuracy and rotation robustness. These results show that a lightweight ($<2$M parameters) equivariant framework can deliver robust anatomical mesh segmentation across diverse label types.
Abstract:Anatomical mesh segmentation requires models that operate directly on irregular surface geometry while remaining robust to changes in coordinate pose across meshes of varying resolution. Existing task-specific mesh and point-cloud methods are not equivariant, and can degrade sharply under test-time perturbation, for example dropping by 28-30 IoU points on 3D-IOSSeg at $40^\circ$ rotation even with matched training augmentation. We present EAMS, an Equivariant Anatomical Mesh Segmentor built on Equivariant Mesh Neural Networks (EMNN), and evaluate it in four dataset settings across three clinical application areas, spanning edge-, vertex-, and face-level supervision. We combine intrinsic mesh descriptors with anatomy-aware priors, including PCA-derived frames for dental arches and liver surfaces, and augment message passing to provide lightweight global context. Across intracranial aneurysm and intraoral segmentation, EAMS variants are competitive with specialized baselines on unperturbed inputs while remaining stable under geometric perturbations, and on liver surfaces they expose a favorable trade-off between canonical-pose accuracy and rotation robustness. These results show that a lightweight ($<2$M parameters) equivariant framework can deliver robust anatomical mesh segmentation across diverse label types.
Accepted as a conference paper to NeurIPS 2026. 30 pages, 7 figures, 21 tables
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2605.08172 [cs.CV]
(or arXiv:2605.08172v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2605.08172 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 286 words
Related coverage
CarveMix-RC: Addressing Rare-Class Imbalance Through Lesion-Aware Synthetic Augmentation for Brain Metastasis Segmentation
arXiv · Sep 29, 2026
Dual-Domain Equivariant Generative Adversarial Network for Multimodal CT-PET Synthesis
arXiv · Jun 12, 2026
When Does Equivariance Help? Canonical Alignment in Neural Fluid Surrogates
arXiv · Sep 30, 2026
Synthesize Realistic 3D Medical Images at Scale to Ship Pre‑Trained Models
NVIDIA · May 22, 2026