Features extraction

Good morning Pyradiomics community.
Is it possible to extract characteristics from an image without informing the mask (region of interest)? that is, I want to extract characteristics from the whole image.


Try using boxes based method

This depends on what you want to acchieve, Currently PyRadiomics requires you to provide a mask, always. However, if you want to extract from the whole image, you can easily generate a ‘full’ mask in python:

import SimpleITK as sitk
import numpy as np

im = sitk.ReadImage('path/to/image.nrrd')
ma_arr = np.ones(im.GetSize()[::-1])  # reverse the order as image is xyz, array is zyx
ma = sitk.GetImageFromArray(ma_arr)
ma.CopyInformation(im)  # Copy geometric info

from radiomics.featureextractor import RadiomicsFeatureExtractor

extractor = RadiomicsFeatureExtractor('path/to/params.yml')
features = extractor.execute(im, ma)

Be aware that this gives features about the texture of the entire image! i.e. a single value per image, for the entire image. If you want more local information, try using voxel-based radiomics:

extractor.execute(im, ma, voxelBased=True)

Be aware that this process will take some time as features are calculated for each voxel!

Hello every body

I need to provide glcm feature maps in different directions. using the guideline I could provide the maps but it gives me just one map per feature, which I think is the weighted average of different directions.
I would appreciate any help to figure it out.

Hi Zahra,
what you want is possible, but requires you to deep-dive into the code.
Basically, by default, separate GLCM matrices are calculated for the different directions. You can access them in the numpy array P_glcm in the GLCM feature class. When features are calculated, they are calculated separately on each glcm matrix, and just before returning, a mean is taken over the different directions (last line in the feature functions). You could implent your own version of a GLCM feature class by copying the original one and modifying this last line.

Thank you very much for your reply. For now I used Scikit-image library, however it does not calculate all the features. So, I did the same thing (access to the matrix to use my calculation) there. But I think pyradiomic is much faster and it is better if can do the same thing using pyradiomic. Thanks again

Hi, sorry to bother you. But I have tried the ‘full’ mask in PyRadiomics and there exists an error: No labels found in this mask (i.e. nothing is segmented)!.

Then, I check the source code of PyRadiomics and the following codes means that there will be an error when there is only one kind of label in the mask file.

if len(labels) == 1 and labels[0] == 0:
raise ValueError(‘No labels found in this mask (i.e. nothing is segmented)!’)

I wonder how to deal with such problem. Can you give me some idea? Thanks for your kindness!

That check is telling you that there are no non-zero values in your mask and that you need to create a segmentation.