Using Masks for Segmentation Tasks

DICaugment not only supports augmenting 3D images but also provides functionality to augment corresponding segmentation masks with the same set of augmentations.

This guide will walk you through the process of using DICaugment to augment 3D image masks in sync with the image transformations.

  1. Import the necessary modules:

    import dicaugment as dca
import nibabel as nib

  2. Read in scan and mask from disk

    scan = nib.load("path/to/scan.nii.gz").get_fdata()

    You can read in the mask just as you would read in a scan. The height, width, and depth of the mask must match the respective height, width, and depth of the image.

    mask = nib.load("path/to/mask.nii.gz").get_fdata()

    For instance segmentation, multiples masks may be needed. The resulting masks should be wrapped in a list

    mask1 = nib.load("path/to/mask1.nii.gz").get_fdata()
mask2 = nib.load("path/to/mask2.nii.gz").get_fdata()
mask3 = nib.load("path/to/mask3.nii.gz").get_fdata()
masks = [mask1, mask2, mask3]

  3. Define an augmentation pipeline using A.Compose:

    transform = dca.Compose([
    # Add your desired augmentation techniques here
    # For example:
    dca.Rotate(p=0.5, limit=20, interpolation=1),
    dca.RandomCrop(height=64, width=64, depth=64)
])

  4. Apply the transformation to your 3D image data and mask

    With a single mask and scan passing through the pipeline, transform must be called using the explicit keyword arguements: image and mask, where the scan should be passed in image and the mask should be passed in mask. The output of this transformation will be a dictionary that contains the augmented scan under the key image and augmented mask under the key mask

    transformed_output = transform(image=scan, mask=mask)
transformed_scan = transformed_output["image"]
transformed_mask = transformed_output["mask"]
    _images/SegmentationExample.png

    If there is more than one mask that are associated with a single scan, you should use the masks argument instead of mask where masks is a list of of individual masks.

    transformed_output = transform(image=scan, masks=masks)
transformed_image = transformed_output['image']
transformed_masks = transformed_output['masks']

You have learned how to use DICaugment to augment 3D image masks for segmentation tasks. Feel free to explore the wide range of augmentation techniques available in DICaugment to further enhance your segmentation tasks. For a comprehensive list of available techniques and their parameters, please refer to the API Reference. If you encounter any issues or have questions, please seek help from the dicaugment community on the DICaugment GitHub Discussions page.

Happy augmenting with DICaugment in your object detection pipelines!