On Thu, 12 May 2005, Fornito, Alexander wrote:
So generally speaking, the thickness values obtained before and after manual editing should be quite similar across the brain. How smooth should the surface be before you can move on? I tend to get slight "protrusions" here and there across the surface and it generally appears to be smooth. Is it just a matter of checking the euler numbers and being happy with the white and pial surfaces when overlaid on the T1? Also, beyond "fix topology correction", what additional processing is done when running "create final surfaces"? Sorry to labour the points, just trying to get an idea of what is/isn't acceptable. Many thanks, Alex
-----Original Message----- From: Bruce Fischl [mailto:[EMAIL PROTECTED] Sent: Wed 5/11/2005 11:30 PM To: Fornito, Alexander Cc: Evelina Busa; freesurfer@nmr.mgh.harvard.edu Subject: RE: [Freesurfer] Manual editing tips?
Hi Alex,
the thickness is the distance between the ?h.white and ?h.pial surfaces. These are initialized with the wm volume, so if it is too far off, they won't converge to the right answer. It's pretty robust, but if you're missing 5-6mm of wm at the crown of a gyrus for example, it probably won't recover the entire thing.
cheers, Bruce On Wed, 11 May 2005, Fornito, Alexander wrote:
> Hi, > Just a bit confused re: Evelina's comment: > > " The surfaces are generated using the normalized brain volume and not > strictly the segmented white matter volume, so edits on the WM volume do > not arbitrarily affect the cortical thickness measures." > > And Bruce's comment: > > " the manual editing can certainly effect the final surface placement. > Mostly if a large piece of wm is missed, or sometimes we edit the brain > volume > directly to remove some dura that gets kept within the pial surface." > > My questions are: > - What surfaces (and at what point) is the thickness calculated from? Is > it from the edited wm and pial surfaces used when create final surfaces > is run? > - To what degree do variations in manual editing affect thickness > estimates? > - Would it be advisable to perform a reliability study to make sure the > manual editing process does not affect thickness estimates too much? > > Thanks again, > Alex > > -----Original Message----- > From: Bruce Fischl [mailto:[EMAIL PROTECTED] > Sent: Monday, May 09, 2005 10:39 AM > To: Fornito, Alexander > Cc: Evelina Busa; freesurfer@nmr.mgh.harvard.edu > Subject: RE: [Freesurfer] Manual editing tips? > > Hi Alex, > > the manual editing can certainly effect the final surface placement. > Mostly > if a large piece of wm is missed, or sometimes we edit the brain volume > directly to remove some dura that gets kept within the pial surface. > > The qsphere is used to guide the topology correction, but itself is not > corrected, so it won't have an Euler # of 2. You can't run > mris_euler_number on thickness, since it's not a surface (but a scalar > field over the surface). You can run it on the ?h.white and ?h.pial > surfaces if you want. > > There is actually a means for manual intervention in the spherical > morph, > but it's rarely needed. > > cheers, > Bruce > > On Mon, > 9 May 2005, Fornito, Alexander wrote: > >> Hi Evelina, >> Let's see if I understand you... >> The pial and white boundaries are calculated on the intensity > normalized/motion/corrected/averaged image, irrespective of manual > editing. >> Then the manual editing is only done to obtain a surface > representation that is visually accurate for display of inflated and/or > flattened surfaces, but has not effect whatsoever on the surface > estimation used for thickness and curvature calculations? >> How about inter-subject registration? Is it affected by manual > editing? >> I've been having problems with my surfaces and am trying to work out > what's going wrong. My euler numbers are 2 for the white and pial > surfaces, but I get the following message when I comupte it for the > qsphere (for one rh.qsphere case): >> >> euler # = v-e+f = 2g-2: 156076 - 468733 + 312496 = -161 --> 81 holes >> F =2V-4: 312496 != 312152-4 (-348) >> 2E=3F: 937466 != 937488 (-22) >> >> total defect index = 185 >> >> Also, I get a "segmentation fault" when I run mris_anatomical_stats rh > thickness for this same person.Visually, I've noticed that the pial and > wm surfaces intersects on some parts of the brain? Could this be the > problem? >> Than ks again for your help, >> Alex >> >> -----Original Message----- >> From: Evelina Busa [mailto:[EMAIL PROTECTED] >> Sent: Sat 5/7/2005 6:51 AM >> To: Fornito, Alexander >> Cc: >> Subject: RE: [Freesurfer] Manual editing tips? >> >> Hi Alex, >> >> The normalized brain volume is created quite early on in the process > -- >> first the raw data are converted from their native scanner format, > then >> motion corrected into what we call the 'orig' volume, then the >> intensities are normalised to what we call the T1 volume, then the > non-brain >> tissues are stripped from the T1 and we have the 'brain' volume. >> >> The WM volume is then segmented out of the brain volume, and that's > what >> is edited, for the purpose of getting the surface topologically > correct. >> So, although it's mostly true that the brain volume is what is used to >> calculate the pial/white boundary and cortical thicknesses (thus >> arbitrary edits to the WM volume won't change that), if the initial >> surface (which is defined by the wm edits) is far enough off, it won't >> find the correct location during the final surface deformation. >> >> In other words, it's important that the wm edits be accurate, but rest >> assured that your edits are not what Freesurfer ultimately uses to >> determine the cortical surfaces. >> >> This should be welcomed as good news! ;) >> >> Good luck! >> >> On Thu, 5 May 2005, Fornito, Alexander wrote: >> >>> Hi once again, >>> Just wanted to also clarify your response: >>> "The surfaces are generated using the normalized brain volume and not >>> strictly the segmented white matter volume, so edits on the WM volume > do >>> not arbitrarily affect the cortical thickness measures." >>> This has me a bit confused. What exactly is the normalized brain > volume (ie., created at which step/which file name?), and if edits on > the WM volume don't affect thickness (and presumably therefore, surface > estimation) why do we edit them at all? There's something I a missing, > and I'd greatly appreciate it if you could point me in the right > direction!! >>> Thanks again, >>> Alex >>> >>> >>> -----Original Message----- >>> From: Evelina Busa [mailto:[EMAIL PROTECTED] >>> Sent: Thu 5/5/2005 4:23 AM >>> To: Fornito, Alexander >>> Cc: >>> Subject: RE: [Freesurfer] Manual editing tips? >>> >>> Hi again, >>> >>> The posterior portion connects to calcarine which makes the big > defect. >>> The tutorial pictures and instructions for filling in the posterior >>> ventricles may give the impression that *any* apparent "hole" in the >>> WM volume will necessarily result in a topological defect, but that's > not >>> true. It depends on whether that hole when looked at in 3D will > still be >>> a hole. This is kind of hard to gauge when going through the slices > but >>> the inflated surface tells the story. >>> >>> Good luck and sorry for my delayed response! >>> >>> On Wed, 4 May 2005, Fornito, Alexander wrote: >>> >>>> Thanks for your comprehensive response! Very helpful! >>>> Just wanted to clarify, why is it that the posterior portion of the > lateral ventricles, and not the anterior portions, result in topological > defects, if both appear as 'islands' inside the white matter? >>>> Thanks again, >>>> Alex >>>> >>>> -----Original Message----- >>>> From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] >>>> Sent: Wed 5/4/2005 2:06 PM >>>> To: Fornito, Alexander >>>> Cc: freesurfer@nmr.mgh.harvard.edu >>>> Subject: Re: [Freesurfer] Manual editing tips? >>>> >>>> >>>> >>>> >>>> Hello Alex, >>>> >>>> A helpful approach when editing the lateral ventricles is to > view the >>>> sagittal plane; usually the filling should not extend anterior > to the >>>> point where the posterior fornix is erased. The axial plane is > also >>>> useful. The tutorial focuses on the coronal plane, which isn't > optimal >>>> for determining when to stop the edits (moving anteriorly). >>>> >>>> Smoothness of the surface -- the aim of manual editing is > chiefly to >>>> eliminate the defects that the subsequent automated topology > fixer >>>> cannot likely handle, whether or not they are easily visible on > the >>>> surface. The large defects outlined in the manual are pretty > safe bets >>>> for editing in this respect, and there are other anatomical >>>> idiosyncrasies that you'll find in normal or other populations > which you >>>> can anticipate having to edit in order to "force" a > topologically >>>> correct surface, e.g. brain lesions resulting in large gaps of > white >>>> matter voxels. >>>> >>>> It's hard to predict what will need to be edited without having > done >>>> quite a few brains, but in general your choice is to either: a) > err on >>>> the side of caution and edit whatever corresponds with a visible > defect >>>> on the inflated surface, or; b) run the topology fixer first > to see >>>> how/whether it could handle the remaining visible defects, with > the risk >>>> that you might need to edit the extra defects afterward, anyway. > And of >>>> course the Euler number calculation can give you precise > information >>>> about a hemisphere's topological defect index. I'd recommend > routinely >>>> editing the standard areas that lead to topological defects (as > per the >>>> tutorial), then running the automated topology fixer, unless it > is a >>>> structurally unusual brain with quite large visible defects. > The >>>> automated topology fixer by and large works, and it's usually a > better >>>> use of operator time to wait on this 5-hour process rather than > spend an >>>> hour or so, perhaps pointlesssly, doing extra edits first. >>>> >>>> The surfaces are generated using the normalized brain volume and > not >>>> strictly the segmented white matter volume, so edits on the WM > volume do >>>> not arbitrarily affect the cortical thickness measures. >>>> >>>> Hope this helps, if not, ask more - >>>> >>>>> Hi all, >>>>> I've noticed that the tutorial's instructions for manual editing of >>>> the ventricles stops at about slice 92 (for Bert's data). Does > this mean >>>> only the posterior portion of the lateral ventricles need to be > filled, >>>> and that there is no need to continue to fill in the ventricles > up to >>>> the anterior horns as well? >>>>> Also, a couple more queries re: manual editing: >>>>> - Sometimes I can see handles emanating from the surface that are > not >>>> in the areas listed in the manual. Are there any guidelines re: > what >>>> should be deleted as opposed to what should be filled? >>>>> - Are there any guidelines for just how smooth the brain surface >>>> should be before moving on to create the final surface? Some of > my >>>> surfaces appear quite smooth (ie., no obvious handles), although >>>> somewhat bumpy, esp. around the posterior-dorsal areas. >>>>> - Am I correct in assuming that what is tissue is edited or not > will >>>> affect surface generation and hence estimates of cortical > thickness? In >>>> this case, is it advisable to carry out some kind of reliability >>>> analysis for morphmoteric studies? >>>>> >>>>> Thanks for your help, >>>>> Alex >>>>> >>>>> _______________________________________________ >>>>> Freesurfer mailing list >>>>> Freesurfer@nmr.mgh.harvard.edu >>>>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer >>>>> >>>> >>>> >>>> >>>> >>>> >>>> >>>> >>> >>> >> >> >
_______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
