Jean-

Thanks for confirming that your software was ignoring the scale slope and intercept factors. For others who have similar issues, dcm2niix is a lossless converter.So if  the raw DICOM data is stored as 16-bit integers, the NIfTI data will be stored as 16-bit integers. The DICOM header values rescale intercept (0028,1052) and rescale slope (0028,1053) will be saved as the NIfTI header values scl_inter and scl_slope. Software that reads the image should apply these scaling factors to the image data (unless the image data is 64 bit float). 

If you have software that ignores these values, you can apply them by using
 fslmaths in.nii.gz -add 0.0 out.nii.gz
Since the default output format of fslmaths is 32-bit float, it will apply the scale factors to the image data. However, the resulting images will require more disk space and be slower to read and save. The influence of this conversion would look like this:


fslhd in.nii.gz
...
data_type INT16
...
scl_slope 1.000000
scl_inter -1024.000000
fslhd out.nii.gz
...
data_type FLOAT32
...
scl_slope 1.000000
scl_inter 0.000000