Import annotations from NIfTI files


Ensure your images and videos have been successfully uploaded and processed

Before you import annotations, make sure your target images and videos are marked as New, or have been previously annotated already.

Once you've added volumetric data to a dataset in V7, you can use CLI commands to import existing annotations from NIfTI files.

You'll need to create a .json file which contains the paths to the image files (these need to match the filenames in the target dataset exactly), the paths to the label files (these are .nii or .nii.gz files with integer values representing classes), a class mapping (a dictionary mapping integer values to class names you want to appear on V7), a mode (this is a string with video, instances or mask)

Here is an example of what such a upload.json file should look like

    "data": [
            "image": "FLAIR.nii",
            "label": "labels/BRATS_001.nii.gz",
            "class_map": {
                "0": "background",
                "1": "oedema",
                "2": "non-enhancing tumour",
                "3": "enhancing tumour",
            "mode": "video"

The image path needs to match the filename used in the Dataset exactly, and the label path is a relative path from where you are running the darwin dataset import command.


Multiple NIfTI uploads at once

You can add NIfTI labels for multiple files in a dataset with a single call to darwin dataset import, just add more entries in the data list with more paths and class maps.

If you set mode to video then each class will be a single video annotation. This makes sense for contiguous 3D objects like a whole organ or a single tumour. If however you are dealing with a single class that is spread in separate instances (e.g white matter lesions in the brain) then you might prefer to set mode: instances to have a separate video annotation for each lesion. The last option mask will write to the raster layer and create mask classes. For more information on masks please visit -

You can now run the only CLI command needed for upload

darwin dataset import your-team/your-dataset nifti upload.json

If the class names supplied do not match ones created in your team the CLI will ask you whether you would like to create those classes.

That's it! you should now be able to see your 3D annotations on the platform.