/deid-dicom

A short script to de-identify DICOM images using `pydicom` and `deid` package

Primary LanguagePython

A short script to de-identify DICOM images using pydicom and deid package

System requirement

  • python (>= 3.6)
  • matplotlib
  • pydicom
  • numpy
  • deid

Installation guide

If you have R (>= 3.6) installed, skip to Section "Run the app from R"

For Windows users, please:

  1. Download and install python3 first, make sure python is added to your PATH environment
  2. Once python 3 is installed, use pip to install the other four libraries
python.exe -m pip install matplotlib pydicom numpy deid

For Linux/OSX users, py3 is natively installed, run these in terminals:

python3 -m pip install matplotlib pydicom numpy deid

If failed, either create a virtual environment or install these packages by adding --user option:

python3 -m pip install --user matplotlib pydicom numpy deid

Notice: deid package depends on an old version of pydicom library (1.3.0). The version is too old to convert colorspace from YBR_FULL_422 to RGB. Please use the following command to check pydicom version. If the version is too low, update it.

The code to check pydicom version: (Please go to terminal/cmd.exe, enter python, hit return button to enter python mode, then run the following code)

import pydicom
# Check if it's less than 1.4.2, if yes, need to upgrade
pydicom.__version__
exit()

Code to update pydicom: in the terminal environment, type:

python3 -m pip install -U pydicom

If you use --user in the previous steps, add to the command before -U option: python3 -m pip install --user -U pydicom.

Ignore the error message.

Run the app directly using Python

  1. Check here for zipped apps, or here if the previous link does not work. Download, extract to some directory (say desktop).

  2. Copy-paste your DICOM images to dicom/ folder (under deid-dicom-0.3/app/)

  3. Remove folder deid-dicom/ if there the folder exists (IMPORTANT)

  4. Edit deid_recipe_data.txt:

  • mask_*: the percentage of images to be masked out. Sometimes the DICOM pixel data has patient information. You might want to mask these pixels out. By default, the mask will be 0.05% in height and 0.3% in width. This means if your image size is 500 x 600, then the top area of size 25 x 180 will be masked out
  • patient_id: current patient ID
  1. Run dicom-deid.py. For Linux/OSX users, you should be able to double-click (or right-click and open) the script. In case the script is not executable, or your operating system is Windows, open terminal, cd all the way to the app directory, type:
python ./dicom-deid.py

(python might be python3 or python.exe, depends on where your python path is)

  1. You should be able to see folder deid-dicom/ created. The de-identified images will be in this folder. There will be png files that display the snapshots (usually useful to adjust masks)

Run the app from R

If you have R installed, the following commands will

  • Download python and install dependencies
  • Download the script app to your desktop
  • De-identify the images
  1. One-time set up: (only run once and no need to run in the future)
# Install packages 
install.packages("remotes")
remotes::install_github("ravepy")

# Configure miniconda environment
ravepy::configure(packages = c('matplotlib', 'pydicom', 'numpy', 'deid'))
  1. Download the latest version (v0.3 as of 2021-06-16)
app_dir <- "~/Desktop/DICOM-app"
url <- "https://github.com/dipterix/deid-dicom/archive/refs/tags/v0.3.zip"

# Download the script to Desktop
tmpfile <- tempfile(fileext = ".zip")
utils::download.file(url, destfile = tmpfile)
unzip(tmpfile, exdir = app_dir)

  1. Please go to your desktop > DICOM-app > deid-dicom-0.3, copy-paste the DICOM images to be de-identified to dicom/ folder.

  2. Go back to R, run the following commands. To make sure correct python is loaded, restart R session first. Run the following commands

work_dir <- "~/Desktop/DICOM-app/deid-dicom-0.3/app"

output_dir <- file.path(work_dir, "deid-dicom")
if(dir.exists(output_dir)){ unlink(output_dir, recursive = TRUE) }
ravepy:::run_script('dicom-deid.py', work_dir = work_dir)