mloecher / tag_tracking

ML based Motion Tracking and Synthetic MR Motion Image Generator

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

tag_tracking

ML based Motion Tracking and Synthetic MR Motion Image Generator

Summary

tagsim/ contains code used to generate motion images, and perform a Bloch simulation to create MR images with proper contrast/features.

torch_tag/ contains the pyTorch implementation of the tracking network and code used to train the network.

tagsim

Software to generate pseudo random deformation MR images. Includes full image deformations and cardiac-like deformations, as well as a GPU accelerated Bloch simulator to generate MR images.

Demo Jupyter notebooks are in the tagsim/notebooks folder.

Installation: This code requires a C library to be built for gridding. Running python setup.py build_ext --inplace in the tagsim folder should build everything. If you are using XCode to on Mac for C compiling, replace setup.py with setup_xcode.py (this disables openMP because stock Mac XCode doesn't support it).

torch_track

Software containing the neural network for tracking MR images. The full network implementation and pre-trained network are included, as well as a demo of its usage on an example dataset. Machine learning is implemented with pyTorch.

Demo Jupyter notebooks are in the torch_track/notebooks folder.

A pre-train network for grid-tagged tracking is in the tagtorch_tracksim/network_saves folder.

Installation: No special installation is required for this software, other than installing required dependencies as they come up.

About

ML based Motion Tracking and Synthetic MR Motion Image Generator


Languages

Language:Jupyter Notebook 98.8%Language:Python 0.9%Language:Cuda 0.1%Language:C++ 0.1%Language:Cython 0.1%