Pixel Scrubbing

Note

Currently only DICOM is supported for pixel scrubbing. If you want to use the pixel scrubbing capability, you need to make sure that the pydicom deid package is installed in your environment. This can be done by running pip install deid or pip install flywheel-sdk[pixel]

Often there will be burned in PHI in the pixels of an image, i.e. Patient Name, DOB, MRN number. This section of the documentation covers how the migration-toolkit can be used to black out or keep a section of pixels

DICOM

DICOM pixel scrubbing is done via the pydicom deid module.

To enable pixel alteration, the top level key alter_pixels must be used. Under the alter_pixels key, there are two setting keys that are relavent: detect and pixels.

Detect

detect uses the built-in rules from pydicom to flag and remove certain known burned-in PHI. detect accepts a list of strings that correspond to filenames in the above folder from the upstream repository.

Pixels

pixels allows specifying custom regions and conditions to keep or remove PHI. pixels accepts a list of dictionaries with two keys

1. either keep or remove which takes a list of coordinates in pixel space as a value:

keep: [x0, y0, x1, y1]

The keep action will remove everything except for the region its value defines, whereas the remove action will keep everything except for the region its value defines.

2. when which takes a string as a value. This string follows the syntax of filters in the deid package

These filters take on the general form of <criteria> <field> <value> and can be chained together with ands (+) or ors (||)

For example, the following filter would match a dicom that has an ImageType with the value SAVE, a SeriesDescription with SAVE in the string, and a value of YES for the BurnedInAnnotation tag :

contains ImageType SAVE
+ contains SeriesDescription SAVE
+ contains BurnedInAnnotation YES

See more on the details of the syntax in the deid package: recipe filters.

Example

Warning

Configuration of pixel-scrubbing including conditions and coordinates is up to the user. The included rules here are NOT thorough.

dicom:
  alter_pixels:
    detect:
      - dicom
      - dicom.xray.chest

    pixels:
      - keep: [0, 0, 10, 10]
        when: |
          contains Modality CT
            + contains Manufacturer GE

      - remove: [20, 20, 30, 30]
        when: |
          contains SeriesDescription Dose Report