Skip to content

tdiepenb/MedicalImageRegistration

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

70 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Medical Image Registration Project

This is the final project for the Computer Vision Course Summer 2024 at University of Cologne. The goal of the project was to implement a non-rigid image registration network.

Single- vs Multiorgan Image Registration

In this project we experimented with the AbdomenCTCT dataset from the Learn2Reg 2020 Task Challenge. We explored if splitting the image using a predefined segmentation and learning individual models for each segmentation could improve registration performance.

For this we used the Registration Framework VoxelMorph to train models for each organ segmented in the AbdomenCTCT dataset and combined these individual registrations to a final registration. We then compared this registration to a VoxelMorph model that uses all organs as input instead of single organs, and to a registration performed using the ANTsPy framework.

Folder structure

The AbdomenCTCT directory contains all notebooks relevant for the final project.

  • The ..._train.ipynb notebooks contain the code used to train the VoxelMorph models.
  • The ...:test.ipynb notebooks contain the code used to evaluate the various models.
  • The AbdomenCTCT_ANTsPy.ipynb notebook contains the code that was used to evaluate the ANTsPy registration
  • The models/ directory contains the weights of the trained models.

The others/ directory contains notebooks that were created to explore other models or datasets but were not relevant for the final project Hand In

Requirements

We used the VoxelMorph Framework at version 0.2. which requires TensorFlow Version <2.15. We therefore used Python Version 3.9.19 or 3.10 with TensorFlow Version 2.15.

The ANTsPy Notebook uses AntsPy version 0.5.3 and was run on Python Version 3.12. as it wouldn't work with Python 3.10. The ANTsPy Evaluation takes quite a bit of time when Run on Colab, we therefore recommend running this Notebook on your local machine.

Other libraries used in the AbdomenCTCT directory included:

To install all libraries/frameworks you can run the following command

pip install jupyter voxelmorph neurite tensorflow==2.15 nibabel matplotlib numpy scikit-learn tqdm antspyx

For the notebooks contained in the others/ directory it might be necessary to install further libraries e.g. torch or natsort

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published