Diff-Def: Diffusion-Generated Deformation Fields for Conditional Atlases
Core Concepts
Diffusion models are utilized to generate deformation fields for conditional atlases, ensuring anatomical fidelity and avoiding hallucinations.
Abstract
Introduction:
Anatomical atlases represent population anatomy.
Conditional atlases target specific sub-populations based on characteristics like age or sex.
Methods:
Diffusion models used to generate interpretable deformation fields.
Morphology-preserving module ensures smooth deformation alignment with attribute-specific images.
Experimental Setup:
5000 brain and whole-body MR images from UK Biobank dataset used.
Proposed method outperforms baselines in generating realistic atlases.
Results and Discussion:
Method demonstrates structural and perceptual superiority over other generative models.
Applicability extends beyond brain datasets to whole-body imaging.
Customize Summary
Rewrite with AI
Generate Citations
Translate Source
To Another Language
Generate MindMap
from source content
Visit Source
arxiv.org
Diff-Def
Stats
Existing approaches use either registration-based methods or generative models [10,12,20].
Our method generates highly realistic atlases with smooth transformations [15].