fix: Fix variable normalization of images

MATLAB does not accept to threshold an image with a negative value, as
is the case when using HU scale. For that reason, the intensity values
of the input image used for muscle degeneration measurement, as well as
the segmentation thresholds expressed in HU, must be mapped to a
'positive' colormap ([0 1] in our case).This normalization was done
until now by mapping the min and max intensity values of the current
image to [0 1]. However, the max intensity value can differ from images
to images, which in turn impacts the thresholds and would prevent
comparison between different patients. For this reason we changed this
'variable' mapping to a 'fixed' mapping, identical for all patients or
images. CT scanners usually have a CT range from -1000 HU to 1000 HU or
-2000 HU to 2000 HU for modern scanners. In our case, we're not using HU
values lower than -1000 and higher than +1000 so all pixel intensity
values of the input image < -1000 HU are set to -1000 HU, and values
> 1000 HU are set to 1000 HU. Then, the input image is mapped to the
[0 1] range (so that -1000=0 and 1000=1), as are the thresholds for
muscle and osteochondroma.