Feature Extraction depending on image type and intensity range

Hi,

I’m extracting features from MRI images. I do an initial feature extraction, followed by selection and binary classification with various machine learning models.
My images are originally int16, the intensity across the dataset varies between -50 and 10000 and I set a bin width of 8. I get an AUC of 0.89.

However, if, for example, I do some rescaling to [0 1] and then multiply by 255 (double) with a binwidth of 2.6 → AUC: 0.6.

Same data…with some alterations…

How can I solve this? What is the problem? I am using PyRadiomics.