How to de-identify a pydicom.Dataset? - addition of example to docs

[fcossio]

Hi,
I have implemented my own logic to load a pydicom.Dataset instance from a database. I would like to de-identify the instance without having to write it as a file and then read it with deid.

Is there anything similar to

def replace_identifiers(recipe, dataset: pydicom.dataset.Dataset) -> pydicom.dataset.Dataset:
    """de-identify a single pydicom.dataset.Dataset instance"""
    ...

?

[vsoch]

Aside from adding typing to deid here, you should be able to do:

if isinstance(dataset, pydicom.dataset.Dataset):
    replace_identifiers(...)

[vsoch]

Also, typing in and of itself doesn't prevent you from providing the wrong type! E.g.,:

In [1]: def func(name: str):
   ...:     print(name)
   ...: 

In [2]: func(1)
1

[fcossio]

After some more digging through the documentation, I solved my problem with the following:

class DeidDataset:
    def __init__(self, recipe_path: str = None):
        """Deidentify datasets according to vaib recipe

        :param recipe_path: path to the deid recipe
        """
        if recipe_path == None:
            logging.warning(f"DeidDataset using default recipe {default_recipe_path}")
            recipe_path = default_recipe_path
        self.recipe = DeidRecipe(recipe_path)

    def anonymize(self, dataset:pydicom.Dataset) -> pydicom.Dataset:
        """Anonymize a single dicom dataset

        :param dataset: dataset that will be anonymized
        :returns: anonymized dataset
        """
        parser = DicomParser(dataset, self.recipe)
        parser.define('remove_day', self.remove_day)
        parser.define('round_AS_to_nearest_5y', self.round_AS_to_nearest_5y)
        parser.define('round_DS_to_nearest_5', self.round_DS_to_nearest_5)
        parser.define('round_DS_to_nearest_0_05', self.round_DS_to_nearest_0_05)
        parser.parse(strip_sequences=True, remove_private=True)
        return parser.dicom
    ...

Thanks for making this tool available.

[vsoch]

oh that's fantastic! Do you mind if I include with our docs somewhere as an example? Even if we create a gist and then link, I think it might be super helpful for future users.

[fcossio]

Of course! I will be OOO for the next two weeks. If you can wait that time, I will make a proper PR afterwards adding the example to the docs.

[fcossio]

Actually, I just found that this only works for files, there are two lines that must be silenced in order for it to work with a dataset that doesn't come from a file:

deid/deid/dicom/parser.py

Line 114 in 0807f20

self.dicom_file = os.path.abspath(self.dicom.filename)

deid/deid/dicom/parser.py

Line 115 in 0807f20

self.dicom_name = os.path.basename(self.dicom_file)

and the file meta here:

deid/deid/dicom/fields.py

Line 254 in 0807f20

datasets = [dicom, dicom.file_meta]

[vsoch]

Yes of course! When you are back ping me I’d you have questions or want any help.

[fcossio]

I'm back 😄

I will need to expose an argument to be able to silence these two lines.

Actually, I just found that this only works for files, there are two lines that must be silenced in order for it to work with a dataset that doesn't come from a file:

deid/deid/dicom/parser.py

Line 114 in 0807f20

self.dicom_file = os.path.abspath(self.dicom.filename)

deid/deid/dicom/parser.py

Line 115 in 0807f20

self.dicom_name = os.path.basename(self.dicom_file)

I propose adding a boolean from_file argument to the __init__ method of DicomParser and then using if-else statements to silence the lines accordingly.

For the DicomField part that needs to be silenced:

and the file meta here:

deid/deid/dicom/fields.py

Line 254 in 0807f20

datasets = [dicom, dicom.file_meta]

I can add the same argument to this method and skip the dicom.file_meta part accordingly.

deid/deid/dicom/fields.py

Line 240 in 0807f20

def get_fields(dicom, skip=None, expand_sequences=True, seen=None):

To keep the interface intact, the default value for the proposed arguments would be True and only if it is necessary, the user could set it to False when needed.

Does this sound good to you @vsoch ?

[vsoch]

A dataset that doesn’t come from a file - what would it be?

[fcossio]

I am loading a dataset that was stored in a database as json. Therefore it contains no filepath or file_meta.

[vsoch]

Gotcha, ok just make functions to derive both of those items then, and if you cannot set to None, and make sure places that use them also respond appropriately.