ISLES 2022 Sub Acute Stroke#

This project folder contains the configuration files, preprocessing, and visualization scripts for the ISLES2022SubAcuteStroke dataset.

For more information, please refer to https://isles22.grand-challenge.org/dataset/.

Related papers:

https://www.nature.com/articles/s41597-022-01875-5.

Note

When running the preprocessing scripts please make sure you have the following packages installed: argparse, connected-components-3d, json, nibabel, numpy, pathlib, random, simpleitk. All those packages, except the connected-components-3d and simpleitk packages, are installed by default if atommic is installed. To install those two packages, please run the following commands:

pip install -r requirements/requirements-isles22.txt

Visualization An example notebook for visualizing the data is provided in the visualize notebook. You just need to set the path where the dataset is downloaded. You just need to set the path where the dataset is downloaded. The notebook is copied and slightly modified from the original notebook provided by the ISLES challenge.

Preprocessing The preprocessing pipeline is implemented in the preprocess_dataset.sh script, consisting of the following steps: 1. Clip to 0 - max values. 2. Normalize the images to zero mean and unit variance. 3. Resample the FLAIR data to the same resolution as the ADC and DWI data. 4. Stack all modalities (ADC, DWI, FLAIR) into a single 3D volume. 5. Fix the orientation of the images. 6. Updates headers and save to NIfTI format. 7. Split the dataset into training and validation sets.

Training/Testing

Important

The ISLES2022SubAcuteStroke dataset is natively supported in atommic. Therefore, you do not need to create a custom dataset class. You just need to set the dataset_format argument in the configuration file to ISLES2022SubAcuteStroke. For example:

train_ds:
    dataset_format: ISLES2022SubAcuteStroke

validation_ds:
    dataset_format: ISLES2022SubAcuteStroke

test_ds:
    dataset_format: ISLES2022SubAcuteStroke

For training a model, you just need to set up the data and export paths to the configuration file of the model you want to train. In train_ds and validation_ds please set the data_path to the generated json files. In exp_manager please set the exp_dir to the path where you want to save the model checkpoints and tensorboard or wandb logs.

You can train a model with the following command:

atommic run -c /projects/SEG/ISLES2022SubAcuteStroke/conf/train/{model}.yaml

For testing a model, you just need to set up the data and export paths to the configuration file model you want to test. In checkpoint (line 2) set the path the trained model checkpoint and in test_ds please set the data_path. In exp_manager please set the exp_dir to the path where the predictions and logs will be saved.

You can test a model with the following command:

atommic run -c /projects/SEG/ISLES2022SubAcuteStroke/conf/test/{model}.yaml

Note: The default logger is tensorboard.