Dear Christian,
Thank you very much for your helpful reply.
It's good to know that robust statistical findings wouldn't change
dramatically.
Yes, I came across that paper as well, and had been wondering whether
similar reasons are responsible for producing different results across
OS X versions with CAT12.
All the best wishes,
Barbara
On 08/03/2019 16:00, Christian Gaser wrote:
> Dear Barbara,
>
> unfortunately, this might happen and was investigated in that paper for the Freesurfer software:
> https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0038234
>
> The authors also found different results for different OS X versions, probably due to the use of different software libraries. Personally, I haven't investigated the effect of different OS X versions, but can confirm that different Matlab versions and operating systems will cause deviations using CAT12. However, the impact on the resulting statistical analysis is rather subtle. Some clusters slightly change their size and even if some local values - as in your case - are changed, the overall difference is quite small and will not impact your statistical results too much.
>
> Best,
>
> Christian
>
> On Thu, 7 Mar 2019 13:29:51 +0100, Barbara Kreilkamp <[log in to unmask]> wrote:
>
>> Dear list,
>>
>> I've processed my data using MATLAB 8.6 (2015b), SPM12 version 7219,
>> CAT12 version 1266e on
>> (a) iMAC running OSX Yosemite (10.10.5)
>> (b) MacBook Pro running macOS Mojave (10.14.2)
>>
>> When subtracting my wm*.nii images generated from (a) and (b) which are
>> directly derived from the CAT12 segmentations from the same participant,
>> I receive a difference in the voxels between -0.924100 and 1.030700.
>>
>> Could you please explain this?
>> Thank you very much,
>> best wishes,
>> Barbara
>
|