Thanks for the info.
We weren't worried about it due to the very minimal nature of the observed differences, but were just wondering why it was different.
-Mike
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Mark Jenkinson
Sent: Tuesday, November 13, 2012 4:18 PM
To: [log in to unmask]
Subject: Re: [FSL] mcflirt discrepancy between refvol and reffile
Hi,
The main difference here is that mcflirt initialises the registration of one timepoint (i.e. volume) using the adjacent one. So when it is used with -refvol 9 it registers timepoint 10 to 9, then does 11 to 9, but initialising this with the result of the 10 to 9 registration. It then does 12 to 9, initialised by the 11 to 9 transformation, etc. When you use the -reffile option it just registers each timepoint separately to the reference image. Unfortunately there are some small inaccuracies in the registration, even between exactly the same images, although these inaccuracies are normally very small and not anything to be concerned about. It is something that we will work on improving, but it is unlikely to have a noticeable effect on anything except very marginal results.
All the best,
Mark
On 13 Nov 2012, at 20:58, Mike Angstadt <[log in to unmask]> wrote:
> It seems that refvol and reffile are resulting in slightly different registrations, even if the underlying image is identical for the two methods.
>
> As an example, I issued the following commands (FSL 4.1.8):
>
> First, registering to the 10th volume in my file:
> mcflirt -in test.nii -o A -refvol 9 -plots
>
> Second, again registering to the 10th volume, but this time extracting it and using reffile instead:
> fslsplit test.nii -t
> mcflirt -in test.nii -o B -reffile vol_0009.nii -plots
>
> Both the resulting timeseries files, and the parameter files are slightly different between these two methods, though I would assume they should be identical. Specifically, in the parameter file, none of my 180 timepoints have all six parameters as 0 as they do for the 10th volume in the refvol method.
>
> Is there a reason I'm overlooking that these methods should not produce identical results?
>
> -Mike
>
**********************************************************
Electronic Mail is not secure, may not be read every day, and should not be used for urgent or sensitive issues
|