The x-ray desensitization tool was developed as a means to increase the inflow of x-ray data from hospitals without breaching patient confidentiality. For more information please refer to our blog post.
The code base was developed in conjunction with a Docker container in order to avoid dependecy problems. If you would prefer to not use the Docker container the following are required:
- Tesseract 4.0 (with LSTM)+
- OpenCV 3.2.0+
- Python 3.5.2+
- Python libraries
- numpy
- pandas
- Pillow
- progressbar33
- pytesseract
- python-gflags
- python-Levenshtein
- scikit-image
- scipy
- python-opencv
When using both the Docker image and the plain source code it is necessary to have all the x-rays in a single folder, the code will recreate another folder of processed x-rays with the same subdirectory structure as the input folder.
The Docker container has been built to clone the repository and execute the code when run. A simple bash script, desensitizer.sh, was made to provide an easier interface for entering the path to the input folder and the path to the output folder. So running bash desensitizer.sh and following the on-screen prompts is all that is required.
clean-xrays.py functions as the API, it has the following parameters.
- --inpath: path to input data folder.
- --outpath: path to output data folder.
- --clean: type of image processing to perform.
- [default] inpaint: fill in the image where the text was, it will appear as if the text never existed.
- mask: mask off where the text was, this will appear as black blocks.