The chaos challenge can be found here: https://chaos.grand-challenge.org/

The script build_chaos_dataset.py converts the dataset as given by the organizers to another of medical images. As given, the dataset consists of single dicom files, which have anonimized dicom files, which all have different InstanceUIDs making these hard to read with standard tools. Run the script for the usage.

For the T1 sequence, the separation of the in-phase and out-phase image is done using the echo time from the dicom header. Next to that the masks are given als PNG files. This tool converts these to a new mask format with the following values:

MAPPING = { 0: 'background', 1: 'liver', 2: 'right kidney', 3: 'left kidney', 4: 'spleen' }

The masks are converted in a .nrrd file where 1 is the liver, and 0 the background.

• Python 3.6
• SimpleITK >= 1.2
• tqdm
• Pillow
• numpy