Learning Deformable Image Registration from Optimization: Perspective, Modules, Bilevel Training and Beyond

This is the official Pytorch implementation of "Learning Deformable Image Registration from Optimization: Perspective, Modules, Bilevel Training and Beyond" (IEEE TPAMI 2021).

Prerequisites

• Python 3.6.8+
• Pytorch 0.3.1
• torchvision 0.2.0
• NumPy
• NiBabel

This code has been tested with Pytorch and GTX1080TI GPU.

Inference

python MultiModal/test.py 

A pretrained MultiPropReg model is available in "models/MPR-T1-to-T2atlas.zip".

Training

If you want to train a new MultiPropReg model using your own dataset, please define your own data generator for train_t1atlas.py and perform the following script.

python MultiModal/train_t1atlas.py

Publication

If you find this repository useful, please cite:

• Learning Deformable Image Registration from Optimization: Perspective, Modules, Bilevel Training and Beyond
  R. Liu, Z. Li, X. Fan, C. Zhao, H. Huang and Z. Luo. IEEE TPAMI eprint arXiv:2004.14557

Acknowledgment

Some codes in this repository are modified from VoxelMorph.

Keywords

Diffeomorphic image registration, hyperparameter learning, iterative refinement for registration, convolutional neural networks, alignment