MECLabTUDA/Atlas-Replay

Atlas-Replay -- Continual atlas-based segmentation (WACV 2024)

This repository represents the official PyTorch code base for our WACV 2024 published paper Continual atlas-based segmentation of prostate MRI. Our code exclusively utilizes the PyTorch version of the VoxelMorph framework as its foundation. For more details, please refer to our paper at [insert paper link].

Table Of Contents

1. Introduction
2. Installation
3. How to get started?
4. Data, splits and pre-trained models
5. Citations
6. License

Introduction

This WACV 2024 submission currently includes the following methods for Continual Learning:

• Sequential Training
• Rehearsal Training
• Riemannian Walk
• Elastic Weight Consolidation
• Bias Correction
• Incremental Learning Techniques

Installation

The simplest way to install all dependencies is by using Anaconda:

1. Create a Python 3.9 environment as conda create -n <your_conda_env> python=3.9 and activate it as conda activate <your_conda_env>.
2. Install CUDA and PyTorch through conda with the command specified by PyTorch. The command for Linux was at the time conda install pytorch torchvision cudatoolkit=11.3 -c pytorch. Our code was last tested with version 1.13. Pytorch and TorchVision versions can be specified during the installation as conda install pytorch==<X.X.X> torchvision==<X.X.X> cudatoolkit=<X.X> -c pytorch. Note that the cudatoolkit version should be of the same major version as the CUDA version installed on the machine, e.g. when using CUDA 11.x one should install a cudatoolkit 11.x version, but not a cudatoolkit 10.x version.
3. Navigate to the project root (where setup.py lives).
4. Execute pip install -r requirements.txt to install all required packages.

How to get started?

• Since our code base follows the VoxelMorph Framework, the models with Continual Learning methods are trained in the same fashion.
• If the VoxelMorph U-Net should be used for simple segmentation, please use the --seg flag, an example train script can be found here.
• The easiest way to start is using our train_abstract_*.py python files. For every baseline and Continual Learning method, we provide specific train_abstract_*.py python files, located in the scripts folder.
• The eval folder contains several jupyter notebooks that were used to calculate performance metrics and plots used in our submission.

Data, splits and pre-trained models

• Data: In our paper, we used seven publicly available prostate datasets where inter-subject samples are rigidly aligned using SimpleITK:
  • Multi-site Dataset for Prostate MRI Segmentation
  • Prostate Dataset from the Medical Decathlon Challenge
• Splits: The train and test splits which were used during training for all methods can be found in the misc folder of this repository.
• Models: Our pre-trained models from our submission can be provided by contacting the main author upon request.
• Prototypes: Our generated prototypes along with the preprocessed dataset can be requested per mail.

For more information about Atlas-Replay, please read the following paper:

Ranem, A., González, C., Pinto dos Santos, D., Othman, A., Bucher, A. & Mukhopadhyay, A. (2023).
Continual atlas-based segmentation of prostate MRI.

Citations

If you are using Atlas-Replay or our code base for your article, please cite the following paper:

@article{ranem2023continual,
  title={Continual atlas-based segmentation of prostate MRI},
  author={Ranem, Amin and Gonz{\'a}lez, Camila and Pinto dos Santos,
          Daniel and Othman, Ahmed and Bucher, Andreas and Mukhopadhyay, Anirban}
}

License

Apache License 2.0