FreeSurfer Surface Reconstruction

This guide provides links to FreeSurfer's Recommended Reconstruction guide, along with some notes, examples, and screenshots describing common issues that come up.

Surface Reconstruction

It is generally recommended that you follow FreeSurfers' Recommended Reconstruction as this document is updated as Freesurfer itself is updated.  They are the true experts in this field.  But note the following (updated Feb 2014):

General:

• FreeSurfer seems to be in the middle of updating their wiki and deploying a new tool (freeview).  Because of this, some of the links on the Recommended Reconstruction page do not seem to work.  You may find that you can find the page you are looking for by adding "_tktools" to the end of the URL.  For example, as of 01/31/14, on the Recommended Reconstruction page, the link for FsTutorial/SkullStripFix brings you to:

https://surfer.nmr.mgh.harvard.edu/fswiki/FsTutorial/SkullStripFix

and tells you "This page does not exist yet. You can create a new empty page, or use one of the page templates."  You can find the page you really want by adding "_tktools":

https://surfer.nmr.mgh.harvard.edu/fswiki/FsTutorial/SkullStripFix_tktools

You'll also see, on the invalid page, a big list of pages in their wiki.  You can find what you are looking for by searching for the last part of the URL, and you should find the _tktools version. There are also pages for "_freeview", which you are welcome to familiarize yourself with and then tell the lab whether we should be using that instead.  But most of us are familiar only with the original "tktools" at this point.

Skip Steps 1 and 2 (setup):

• As long as you followed the above Setting Up FreeSurfer instructions (or you are on a CoreB machine and don't have any reason to use a different SUBJECTS_DIR than the default one), we have already setup FreeSurfer and the default SUBJECTS_DIR in our .cshrc file.

Step 3 (initialize and copy T1):

• You should ONLY use the T1 volumes as invol1.  You could supply multiple T1s if you have them (but that is not usually necessary), but DO NOT input your T2s here. In particular, if you provide FreeSurfer with a T1 and a T2 volume, they will be averaged together and you will essentially loose all of your contrast in the averaged image.  T1s only!

Step 4 (run recon-all):

• It is possible to run recon-all in "steps" (recon-all -autorecon1; recon-all -autorecon2; recon-all -autorecon3).  This can be useful in that you don't have to wait until the surface reconstruction (autorecon2) and registration (autorecon3) are complete to check some of the earlier steps.  However, FreeSurfer now recommends simply running all the steps the first time.  Additionally, you may run into an error with -autorecon2 failing if you run it before running all of the steps (I'm not sure how this would occur, but it does.  FreeSurfer is aware, there is an updated file you can download as a fix, and presumably they will fix it in a future release).  Bottom line, if you are not pressed for time, it is probably better to go with FreeSurfer's recommendation as you might run into more headaches doing it step-wise.

Step 5 (check Talairach):

• Do not spend too much time on the Talairach alignment.  Although FreeSurfer uses this in various steps of its pipeline (so make sure it isn't completely wonky), we don't typically use it for group analysis.  Other group analysis options include surface-based registration (done automatically by FreeSurfer), MNI alignment (FSL or AFNI), or if you must, Talairach alignment in other programs (AFNI).

Step 6 (check skull strip):

• Control points are very useful, and you may find that they are needed in the anterior temporal lobes where these can be some signal loss (due to the ear canals) and a lower contrast between gray and white matter.

• I have had some success with the gcut method (look for "Using gcut" on the SkullStripFix page.  It isn't perfect, but it does seem to help.  If your pial surface does not extend beyond the cortex (and into the skull/scalp), then you should be fine for cortical reconstruction.  However, we will create a NoSkull version of the final volume in AFNI/SUMA, and if you have a really clean brainmask.mgz in FreeSurfer, we can simply copy that over.  A really good skull-strip (no skull, no missing brain) is very important for volume registration (functionals to surface anatomy; in-session anatomy to surface anatomy; T2 to T1 used for surface) and for creating the skull and scalp surfaces for a Boundary Element Model (BEM) to be used for source localization of HD-EEG data.

Step 7 (check white and pial surfaces):

• • When checking white matter, be sure to clean up the optic nerve.

  • You may also identify clear veins along different parts of cortex (causing expansion of the pial surface).  These are usually best seen when flipping between slices as a moving bright spot along the outside of cortex or within the sulci. You can remove these if they are clear.

  • Be particularly aware of pial expansion into the cerebellum, as those represent many parts of ventral visual cortex, which we obviously care about.

Step 8 (check segmentation):

• I have never used FreeSurfer's segmentations before (except, to some degree, the surface ones based on the surface registration), so I am not familiar with the ins and outs of the step.  I suppose I usually assume it's all okay.

Other parts of the Freesurfer wiki may be of help.

Next Steps

After creating a new surface, you'll need to Set up the Subject's SUMA Directory.

To Add

• • Some notes on the organization of surfaces, anatomy, etc. on the CoreB computers

  • Additional flags to use for recon-all (e.g., -3T), what they do and why we might consider using them

  • Screenshots showing examples of these common issues following the default preprocessing: