JISCMail - FSL Archives

Hi again!

Just to quickly add that now I am even more confused.

I tried to run the segmentation step on its own (meaning using run_first instead of run_first_all).
I used the flirt matrix output of the run_first_all command that i have used previously.
I tried to segment the right accumbens using the same model as my more recent analysis and 50 modes of variation.
Now the result of the segmentation look different both from my first and my second analysis.
Maybe i don't get the default value used by run_first_all ? isn't 50 the number of modes suggested for the accumbens area ?

Alain

Date: Tue, 21 Jul 2015 10:40:45 +0200
From: [log in to unmask]
Subject: Re: [FSL] first segmentation: same input images different output ?
To: [log in to unmask]

Hi!

Of course I can elaborate. I would be happy to get any help in understanding this!

I used skull stripped images cause I found that for my dataset this worked better.
So as for now i checked all the "in-between step" for the two analysis.
The images that I used as imput to bet are the same, the images output of bet are the same. The results of the first_flirt step also look exactly the same.
I also used exactly the same script for running the analyis, changing only the path to the images.
The only thing that differ is, at least for what I can see, in the final segmentation.
The more recent analysis give clearly larger estimates for all subcortical nuclei except for accumbens, which get "eaten" by putamen and caudate.
If I check the model reported in the bvars file for,e.g., the accumbens, the first analysis report
/usr/share/fsl/5.0/data/first/models_336_bin/R_Accu_bin.bmv
and the more recent one report
/usr/local/fsl//data/first/models_336_bin/R_Accu_bin.bmv
So it seems that also the model used is the same.

At the bottom line, I had some significant result with the first analysis that I don't have any more with the new.
I would just like to understand why!

I cannot attach the images as they are too big for jiscmail, is there a repository where I can upload them for you to take a look ?

Best

Date: Mon, 20 Jul 2015 19:02:29 +0000
From: [log in to unmask]
Subject: Re: [FSL] first segmentation: same input images different output ?
To: [log in to unmask]

Hi,

If you were using the most recent version of FSL in April 2014 then that would be 5.0.6. We are currently at 5.0.8. However, there were no real changes to the way that FIRST works in between these releases, so this would not explain differences that you are seeing. There is also no way of interrogating what precise version of FSL was used to process the files that you have.

So I’m afraid I cannot explain your differences.

I am surprised that any differences are remarkable. Can you elaborate on this?

All the best,

Mark

From: FSL - FMRIB's Software Library <[log in to unmask]> on behalf of Alain Imaging <[log in to unmask]>
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
Date: Monday, 20 July 2015 10:37
To: "[log in to unmask]" <[log in to unmask]>
Subject: Re: [FSL] first segmentation: same input images different output ?

Hi, I bump this question since this is really bugging me

Hi everybody!

I was processing a dataset with FIRST in order to perform shape analysis.

I had already processed a subsample of this dataset some time ago in order to do some preliminary analysis.

I am feeding to FIRST the same images and I am running on the same machine, with the same version of FSL used for the analysis of that subsample.

Surprisingly, if I compare the segmentation performed some time ago and the one performed now, I found some slight but nonetheless remarkable differences in >segmentation.

Is this even possible, or the only explanation is that I have done some mistake ?

Thanks in advance for any tip

Alain

To add some details: I have performed the segmentation of the first subsample around April 2014. I run the analysis on a server, and I am not aware if the administrator has changed the FSL version in between. The model used for the first subsample was nonetheless in a directory called /FSL/5.0/.

Have there been major patches to the FIRST model between April 2014 and now ? Is there a method to know exactly with which version of FSL a FIRST segmentation was performed ?

Looking forward for your answers

Alain