OpenCXR is an open source collection of chest x-ray (CXR) algorithms and utilities maintained by the Diagnostic Image Analysis Group (DIAG) at Radboud University Medical Center, Nijmegen, the Netherlands. www.diagnijmegen.nl

The algorithms currently offered are as follows:

• Image Sorter:

  This algorithm is designed to help users sort out volumes of downloaded CXR data. Given a 2D image, Image Sorter will identify it as one of the following image types:

  • Frontal PA chest X-ray
  • Frontal AP chest X-ray (distinction between PA and AP is not as good as distinction between other types)
  • Lateral chest X-ray
  • Other (not a chest X-ray)

  Image Sorter will additionally provide the following useful information for Chest X-Ray images:

  • Rotation (identifies if the image is rotated 90, 180 or 270 degrees)
  • Inversion (identifies if the X-Ray is stored with bright values representing air-filled regions)
  • Lateral Flip (identifies if a lateral Chest X-Ray is stored with the spine to the left of the image)

  Image Sorter is designed to work on raw data and it is not recommended to perform image normalization/standardization procedures in advance.

• Lung Segmentation:

  Given a PA chest X-ray this algorithm will provide a segmentation of the lung fields as a binary image.
  This algorithm may work on AP images but is not trained or tested on them. Performance may be better on images that have been standardized (see Chest X-ray standardization below )

• Heart Segmentation:

  Given a PA chest X-ray this algorithm will provide a segmentation of the heart as a binary image. This algorithm may work on AP images but is not trained or tested on them. Performance may be better on images that have been standardized (see Chest X-ray standardization below )

• Chest X-ray standardization:

  Given a frontal chest X-ray this algorithm will

  • standardize the image intensities using energy bands and a region of interest (obtained by lung segmentation). This process is described in the following paper: Localized Energy-Based Normalization of Medical Images: Application to Chest Radiography, Philipsen et. al., IEEE Transactions on Medical Imaging, 2015
  • (Optionally) crop the image to the lung bounding box
  • Resize the image to a specified square size (e.g. 1024x1024), preserving aspect ratio and padding the shorter axis.

The utils library contains functions for common or useful functions when working with chest X-rays. These include:

• reading/writing images
• resizing images
• rescaling intensities
• masking/cropping
• rotating/flipping

In general if you work with OpenCXR algorithms it is recommended to use the OpenCXR image reader/writer and other utilities.

• pip (if you just want to use the algorithms/utilities)
• git lfs (only if you want to clone this repository including model weights)

You can get the latest version of OpenCXR by simply running the command below

pip install opencxr

A useful place to get started is to look at the test code for the algorithm you want to run.
You'll find a python file to test each algorithm in the tests folder.
Any algorithm can be loaded by calling my_algorithm = opencxr.load(algorithm_name) . Algorithm names are listed at the top of this file.
The loaded algorithm can then be run by calling my_algorithm.run(required_input).

The code snippet below gives a suggestion of how to get started with the cxr standardization algorithm.

import opencxr
from opencxr.utils.file_io import read_file, write_file

# an algorithm is created with a call to opencxr.load() .  
cxr_std_algorithm = opencxr.load(opencxr.algorithms.cxr_standardize)

# provide the path to the file you want to read in
full_cxr_file_path = input_folder_cxr + '/' + input_cxr_file

# read the  file (supports dcm, mha, mhd, png)
img_np, spacing, dcm_tags = read_file(full_cxr_file_path)

# Do standardization of intensities, cropping to lung bounding box, and resizing to 1024 square
std_img, new_spacing, size_changes = cxr_std_algorithm.run(img_np, spacing)

# write the standardized file to disk
output_cxr_loc = output_folder_cxr + '/' + input_cxr_file
write_file(output_cxr_loc, std_img, new_spacing)