Research map / 3D vision / Face and shape
NePhi: Neural Deformation Fields for Approximately Diffeomorphic Medical Image Registration
Lin Tian, Thomas H Greer, Raul San Jose Estepar, Roni Sengupta, Marc Niethammer
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| represent deformation with a conditioned MLP (NePhi) instead of a voxel field, single-resolution | GradICON-style registration · DirLab (single-resolution) | Peak train memory (MB): 6394 -> 2466 | Improved |
| add instance optimization (200 iterations) to NePhi versus GradICON with 50 (ablation) | NePhi single-resolution · DirLab | mTRE (mm): 5.44 -> 1.73 | Improved |
| use multi-resolution NePhi instead of single-resolution (ablation) | NePhi · DirLab | mTRE (mm): 5.44 -> 2.78 | Improved |
| NePhi multi-resolution compared with multi-resolution GradICON | NePhi · DirLab (multi-resolution) | mTRE (mm): 1.93 -> 2.78 | Hurt |
| use NePhi instead of LapIRN in multi-resolution setting | multi-resolution registration · DirLab | mTRE (mm): 4.24 -> 2.78 | Improved |
| NePhi on brain MRI instead of GradICON | multi-resolution registration · HCP brain MRI | DICE: 78.7 -> 76.2 DICE, memory 6716 -> 1760 MB | Mixed |
| increase training image resolution from 21 to 175 (ablation) | NePhi · DirLab | mTRE (mm): 16.85 -> 5.44 | Improved |
| increase training image resolution from 175 to 256 (ablation) | NePhi · DirLab | mTRE (mm): 5.44 -> 5.30 | No clear effect |
| increase IO iterations from 100 to 1600 (ablation) | NePhi(S) · DirLab | mTRE (mm): 1.94 -> 1.51 | Improved |
| reduce regularizer sampling ratio over a wide range (ablation) | NePhi · Triangles and Circles | Dice and regularity | No clear effect |
| use gradient inverse consistency regularizer instead of diffusion regularizer in NDF instance optimization (ablation) | NePhi · Triangles and Circles hollow | Dice at convergence | Improved |
| apply gradient inverse consistency to voxel-based transformation in optimization (ablation) | voxel deformation field · instance optimization | regularity | Hurt |
Closest papers by meaning
- NODEO: A Neural Ordinary Differential Equation Based Optimization Framework for Deformable Image Registration (CVPR 2022 · similarity 0.89)
- A Physics-Informed Deep Learning Deformable Medical Image Registration Method Based on Neural ODEs (IJCV 2025 · similarity 0.89)
- Learning Deformable Image Registration From Optimization: Perspective, Modules, Bilevel Training and Beyond (TPAMI 2021 · similarity 0.88)
- An Unsupervised Learning Model for Deformable Medical Image Registration (CVPR 2018 · similarity 0.88)
- Deep Learning in Medical Image Registration: Magic or Mirage? (NeurIPS 2024 · similarity 0.85)
- A Plug-and-Play Image Registration Network (ICLR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
