This wiki page contains the students questions and answers from the April 29 2013 Freesurfer course.

Question List

It is easier for me to get a NIFTI file of the T1 structural image than the original DICOM files. I have been converting the NIFTI to a freesurfer format using mri_convert. I can get the original DICOM files if that's better. Is it worth going back to the DICOM if NIFTI are more readily available?


Could you please repeat the pros and cons of different fyletypes for input to recon-all? I believe I heard Doug say he strongly recommended DICOM. Why is this? thank you very much!

The danger of using a different input than DICOM is that sometimes important header information gets lost. This can result in flipped images (you may not even notice a left-right flip) etc. Usually NIFTI should be fine (assuming the header did not get corrupted). MartinReuter

Can you use skull stripped images as input files?

It is not recommended to use skull stripped images, as several procedures assume that the skull is there. The skull also contains important information used during the atlas (Talairach) registration. MartinReuter

Group Analyses: Freesurfer makes use of different scan characteristics to recon surfaces and estimate thickness etc. What if the groups you compare differ on either of these characteristics? For example, like with the Huntington scans you mentioned, what if one of your groups has more movement artifacts? Or with the nudging of the grey-white matter surface, what if this border is (in reality) more nudged in one of your groups than in the other?

If the scans differ with respect to these characteristics, your study will be biased. You will never know if the effect you measure is due to the disease or due to, e.g., motion artifacts. That is why it is extremely important to control these parameters (use the same pillow, place supporting cushion at the side of the head, scan a few times, select the best scan without motion, etc.). There can be other problems, e.g. differences in hydration level. MartinReuter

It can be a problem if the contrast characteristics are different between your groups for reasons other than biology. Motion may cause something like this. There is not a good way to test whether your populations moved differently just from the anatomical. You can look at the amount of signal outside of the head as a measure (motion causes signal outside of the head). Different amount of nudging will occur if the GM/WM contrast is different or the GM/CSF contrast or the thickness are different. If these are happening for biological reasons, then it probably reflects differences in thickness. DougGreve

ROIs for FSL: I want to use FreeSurfer to create subject specific ROI's for subcortical structures. Is there an easy way to convert an ROI from FreeSurfer to work with FSL? I have used the conversion tool to make it a NIFTI, but then I always get an error that the dimensions are not right.

How did you convert the ROI label image? The conversion usually does not change the resolution of the image. You can try to

mri_convert -rl <target_with_correct_dimensions> -it nearest roi.mgz roi-in-new-space.nii.gz

which will reslice the image into the new resolution using nearest neighbor interpolation. This may however affect the ROI labels as the result may look very noisy. If you manually label the image, better do it directly in the target space. Without more information it is difficult to help. Post your question with more details on the FreeSurfer list. MartinReuter

If the dimensions are not right it probably means that you need a spatial transformation. If you want to go back to the native anatomical space instead of the conformed, then see this page DougGreve

Surface-based Volume: Hi, as Doug stated in his talk volume= surface x thickness This might be an easy question but I was which surface is taken for this calculation? White or pial? Or maybe something in between? Depending on the curvation and folding in this reagion this might be making a difference... Thank you!

The white surface is used. DougGreve

Monte Carlo Simulation: Hi, as mentioned in the talk the results of the monte carlo simulation are already precached. Since I used the program alphasim before I know that one of the input parameters for the simulation is the smoothness of the data. So are there precached results for data with different smoothness? Or how do account for the smoothness in the data? Thank you!

The cached simulations are done over a range of smootheness. When you run mri_glmfit, it computes the FWHM of your data from the residuals. When you run mri_glmfit-sim, it reads the fwhm.dat file to determine which data in the simulation table to use. DougGreve

Different HRF for different regions: I have heard about some work suggesting that different HRFs (rather than the canonical HRF) may be more appropriate for some brain regions. Is this something your group has been concerned with? Is there a way to use different HRFs in FS-FAST?

We have not been involved in this. There is not a way to use different HRFs at each voxel with FSFAST. There are some other things that can be done like adding derivatives or using an FIR that will give you some flexiblity, but it is not the same using a different HRF at each voxel. DougGreve

Parameters for e3 mask: I was wondering whether it is possible to set how many points the e3 mask is eroded, e.g. only 2 instead of 3 points. It seems to be hard coded, but it would be useful to be able to set this as a parameter.

I created a new version of mkbrainmask-sess which you can get from here: It has a -ne3erodes parameter which allows you to set the number of erodes. DougGreve

How do you turn on v4 type whitening in v5.x?

Add -fsv3-whiten to your mkanlaysis-sess command line. DougGreve

How do I do RFX and FFX analyses with mri_glmfit?

RFx is the default. For FFX, you add the following to your command line

--yffxvar cesvar.nii --ffxdofdat ffxdof.dat

where cesvar.nii is the variance measure from your first level analysis and ffxdof.dat is a simple text file with the total DOF from all your subjects. Both of these are created by isxconcat-sess. DougGreve

CVS: Why is CVS not applied as the standard registration tool for registration of different subjects before doing (plain, anatomical) group analyses in qdec or glmfit?

For analysis of cortex, you will always want to use surface-based analysis in order to use surface-based smoothing, avoid subcortical structures, and do stats on thickness. For VBM-style analyses, it is possible to use CVS and input the results to mri_glmfit (not possible to use with QDEC for now). DougGreve

FsTutorial/QuestionAnswersApril2013 (last edited 2013-11-03 11:32:23 by AllisonStevens)