kumarshreshtha / image-registration-cnn

A pytorch implementation of CNN based MRI image registration based on Christodoulidis Stergios et al. "Linear and Deformable Image Registration with 3D Convolutional Neural Networks", 2018.

Linear and Deformable Image Registration with 3D CNN

A minimal PyTorch implementation of CNN based MRI image registration from the paper Christodoulidis Stergios et al. "Linear and Deformable Image Registration with 3D Convolutional Neural Networks", 2018.

For a quick introduction to image registration and summary of the paper check out this presentation.

Note: This is not the official implementation. I had to present this paper for an undergrad internship interview back in 2019 and I ended up writing this quick prototype to go with it. Cleaned it up a bit recently but it's still a barebone running prototype than anything else.

Network Architecture

Requirements

You can download all requirements for this project using pip as follows:

pip3 install -r requirements.txt

Usage

Train

As mentioned above, the implementation is pretty barebone. Once you have your data directories set up, you can change the hyperparameters in train.py and simply launch the train script.

python3 train.py