/ISLES22_SEALS

ISLES22 Acute Ischemic Stroke segmentation competition, team SEALS final submission

Primary LanguagePythonApache License 2.0Apache-2.0

Property data preprocessing and nnUNet for efficient Acute Ischemic Stroke segmentation

Introduction

This is the repository of team SEALS submission for ISLES22 competition.

Installation

This repository is based on nnUNet. Please follow the installation instruction of nnUNet. We recommend to install conda environment before installing this repository. We will provide the installation instruction of this repo.

Usage

If you do not want to install conda, please skip to the step 3.(We do not recommend to install nnUNet straightly.)

  1. Let us hypothesis you are using conda environment named "nnunet", activate conda environment "nnunet"
conda activate nnunet
  1. Run the following command to install basic pytorch、torchvision library and corresponding cudatoolkit.
conda install pytorch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0 cudatoolkit=11.3 -c pytorch
  1. Clone this repository.
cd ~
git clone https://github.com/Tabrisrei/ISLES22_SEALS.git
  1. Run the following command to install nnUNet.
cd ISLES22_SEALS
pip install -e .

If successfully installed all required packages, you can run the following command to test your own dataset(You may need to install SimpleITK version under 2.0.2 to avoid some unpredictable errors).

  1. Download zip file of our models from Google Drive and unzip it.
  2. Put the "nnUNet_trained_models" folder you've got into the directory ~/ISLES22_SEALS/data/.
  3. Convert your dwi and adc image to mha format (Do not forget metadata)
  4. Put your dwi and adc data into /input/images/dwi-brain-mri/ and /input/images/adc-brain-mri/ folder respectively. Or you may modify the raw_data_diro in ~/ISLES22_SEALS/nnunet/dataset_conversion/Task500_Ischemic_Stroke_Test.py
  5. Run the following command to generate the results.
bash nnunet_launcher.sh

After the results are generated, you can check the results in /output/images/stroke-lesion-segmentation/ folder.

Questions

Please contact gtabris@buaa.edu.cn

Acknowledgement

  • This code is adapted from nnUNet
  • We thank Dr. Fabian Isensee etc. for their elegant and efficient code base.