BraTS Numpy#

Format specification#

The original BraTS dataset is available here.

Supported annotation types:

Mask
Cuboid3d

Import BraTS Numpy dataset#

A Datumaro project with a BraTS Numpy source can be created in the following way:

datum project create
datum project import --format brats_numpy <path/to/dataset>

It is also possible to import the dataset using Python API:

from datumaro.components.dataset import Dataset

brats_dataset = Dataset.import_from('<path/to/dataset>', 'brats_numpy')

BraTS Numpy dataset directory should have the following structure:

dataset/
├── <img1>_data_cropped.npy
├── <img1>_label_cropped.npy
├── <img2>_data_cropped.npy
├── <img2>_label_cropped.npy
├── ...
├── labels
├── val_brain_bbox.p
└── val_ids.p

The data in Datumaro is stored as multi-frame images (set of 2D images). Annotated images are stored as masks for each 2d image separately with an image_id attribute.

Export to other formats#

Datumaro can convert a BraTS Numpy dataset into any other format Datumaro supports. To get the expected result, convert the dataset to a format that supports segmentation masks or cuboids.

There are several ways to convert a BraTS Numpy dataset to other dataset formats using CLI:

datum project create
datum project import -f brats_numpy <path/to/dataset>
datum project export -f voc -o <output/dir> -- --save-media

datum convert -if brats_numpy -i <path/to/dataset> \
    -f voc -o <output/dir> -- --save-media

Or, using Python API:

from datumaro.components.dataset import Dataset

dataset = Dataset.import_from('<path/to/dataset>', 'brats_numpy')
dataset.export('save_dir', 'voc')

Examples#

Examples of using this format from the code can be found in the format tests