| T O P I C R E V I E W |
| vinod |
Posted - Jul 06 2007 : 5:37:40 PM
I have a general question about FSL registration. I am trying out a few different models for my data analysis using FSL. To facilitate this, I tried to do my preprocessing first and then work on the preprocessed data, so that I save the time needed to do these steps again.
However, when I run my models, it appears as if FSL still performs registration every time I run my model. Initially, it did not make sense to me but after looking up the forum, it appeared as if they intentionally do this to run stats on the raw non-interpolated images and then apply registrations to standard space. I could not find more details on the rationale for this though. Does anyone have any ideas?
Is there anyway I can bypass having to do registration each time I run a model, like I would in spm? Even if it has to register again, given that the images are the same, it should just be application of the transformation parameters which should not take as long as it takes for me currently. |
| 5 L A T E S T R E P L I E S (Newest First) |
| syam.gadde |
Posted - Jul 23 2007 : 5:41:47 PM
I bet you could derive between-run motion information from the info in the "mc" directory and the transformation matrices to standard space in "reg". That's a question probably best addressed to the FSL mailing list (they are quite responsive and helpful).
Yes, each run is individually registered to standard space (by default) before being input into a second-level analysis.
I don't know about your weird activation problem, but it is easy enough to inspect the registration (which is a good idea) by looking at the .gif images in "reg" or by viewing the report.html in a web browser.
|
| vinod |
Posted - Jul 23 2007 : 5:17:58 PM
A quick followup of that earlier question. So that means when you are using FSL, there is no way of knowing how much the subject moved between runs, is that right? Also, technically, subject motion between runs is not as critical when using FSL for analysis as each run is individually registered to the standard space, is that right?
I am actually getting some weird activations for one of my contrasts in the second level (combining runs from one subject) on the ventricles. These don't exist in the first level outputs. So, is it possible that it is a registration issue? |
| syam.gadde |
Posted - Jul 09 2007 : 12:25:05 PM
FSL does not know that your second-level analyses are combining data from the same subject, though that is a common approach. Furthermore, though the data may be from the same subject and the same scanning session, the head may have moved between runs, so it can't make any assumptions that the data is already registered to each other. That's why it needs to perform registration on every first-level analysis... |
| vinod |
Posted - Jul 09 2007 : 12:17:22 PM
Yes, I agree with the rationale Syam. This means that effectively there should be no need to do registration till your final level analysis right? I should technically be working on non-registered data in the first and second levels and have to do the interpolation in the last level when I am combining across subjects? |
| syam.gadde |
Posted - Jul 09 2007 : 11:03:08 AM
I think the rationale for transforming the statistics (rather than the input data) is that it is safer to introduce the interpolation/smoothing inherent to registration after running the statistics, rather than having it affect the initial calculation of the statistics themselves. It seems reasonable to me, but I don't have any data to back it up.
I looked through the code, and I can't find any way to have it do the right thing in the higher-level analyses if you don't let FSL do the registration itself during the first-level analysis, and it doesn't look like there is any clean way to use a pre-existing registration (without changing the FEAT code itself). |
|
|
