This script uses SimpleITK to read medical images and OpenCV for the GUI in order to provide a simple and fast way to crop 3D boxes (for instance to generate ground truth for a classifier). It is a 3D adaptation of [1].

ESC or q : exit
SPACE : save box
ENTER : skip, go to the next file without saving.

If a cropped box is already found, the file is skipped, so to change a cropped box, you need to delete the corresponding png.

A regular expression to parse the png files is:

import re
pattern =  r'(?P<raw_file>[^/]+\.nii)_(?P<z>\d+)_(?P<y>\d+)_(?P<x>\d+)_(?P<d>\d+)_(?P<h>\d+)_(?P<w>\d+)\.png$'
match = re.search( pattern, filename )
m = match.groupdict()

See ./run.sh for an example for cropping, and ./parse.py ground_truth/ boxes.txt for an example for parsing the cropped boxes to a text file.

[1] N. Seo. Imageclipper - A Tool to Crop Images Manually Fast. Imageclipper, 2008. https://code.google.com/p/imageclipper/