Finding the largest physical region (bounding box) across an arbitrary number of input files (SimpleITK)

gdevenyi · August 1, 2022, 5:59pm

I’m working on an arbitrary image averaging script, where I’d like to assume input images are not guaranteed to overlap (completely, or even partially). As such, I need to construct a “common” space which contains all the input images. Finding the voxel dimensions is easy per-image, but before I go about writing something to locate all the “corners” of the data in space manually, is there a more clever solution in ITK I might be able to use?

dchen · August 1, 2022, 6:18pm

I don’t know of anything clever, but I did write the code for computing a world space bounding box of a volume.

Here’s my code to ‘regularize’ a volume. It makes a volume with cubic voxels and an orientation of the identity matrix.

github.com

dave3d/dicom2stl/blob/main/utils/regularize.py

#! /usr/bin/env python

import getopt
import sys
import SimpleITK as sitk

# Regularize a volume.  I.e. resample it so that the voxels are cubic and the
# orientation matrix is identity.
#


def regularize(img, maxdim=-1, verbose=False):
    dims = img.GetSize()

    if verbose:
        print("Input dims:", dims)
        print("Input origin:", img.GetOrigin())
        print("Input direction:", img.GetDirection())

    # corners of the volume in volume space

This file has been truncated. show original

gdevenyi · August 1, 2022, 6:31pm

Thanks for the code, I can definitely steal some of that.

Since you’re isotropizing images, you might find this useful, Resampling to Isotropic: Signal Processing Theory

dchen · August 1, 2022, 7:38pm

Interesting thread. Yeah, I don’t do any sort of filtering, so it can result it artifacts.

gdevenyi · August 2, 2022, 5:49pm

Thanks very much again @dchen, I stole your code mostly wholesale to sort out the bounding box, now I can combine arbitrary-space images into a common “mega” image covering the full bounding box, as I wanted.

dchen · August 2, 2022, 10:52pm

You’re welcome. I’m glad you found my code useful.