 |
 |
Hello - is it possible that some of your data voxels are set to NaN? if you create a nan-mask for one of the problematic images, e.g. fslmaths <input> -nanm <output> , does it contain any non-zero voxels ( check with fslstats -V )? Kind Regards Matthew > On 14 Aug 2016, at 17:19, Michael <[log in to unmask]> wrote: > > Hi, > > It appears that FSL behaves unexpected on some servers, with 0/1 float data being potentially read as NaN. > > On a Centos server, looking at minimum and maximum values of such a "float 32" binary image (fslstats -R my_image.nii.gz) gives -nan -nan. When I copy the same file on my Ubuntu system with the same version of fsl (5.0.9) as the Centos server, I obtain the correct range, 0.000000 1.000000. > > Please note that it was particularly hard to debug the whole workflow to pinpoint this issue in FSL behavior on different OS systems. I assume other users had this issue but could not find the origin of their errors. I could find an unelegant solution which consists in converting the image to int (fslmaths -odt int) [which provided the range -2147483648.000000 1.000000], and then binarizing it, to obtain the correct image (with range 0.000000 1.000000). > > Would someone have more insight and what is actually happening? It seems related to data type reading from different OS but I am not clear of the exact mechanism. > > Many thanks, > > Michael