Thank you Mark.

if I wanted to add SIENAX-derived head size as a covariate for FIRST
corrected output (not vertex results) and use, say, a standard
regression multiple regression model in a statistical package, would I
use subjects normalized BRAIN (surrogate for head size; for example,
calculated below from the practical [1192234.82]) to adjust or control
for head size? That is, BRAIN values could be a covariate/regressor,
along with other covariates, such as Medication and gender, all of
which may contribute some variation to the outcome variable FIRST
subcortical volume.

----------  convert brain volume into normalised volume  --------------

tissue             volume    unnormalised-volume
pgrey              496752.82 454866.29 (peripheral grey)
vcsf               79257.48 72574.44 (ventricular CSF)
GREY               584635.30 535338.46
WHITE              607599.52 556366.32
BRAIN              1192234.82 1091704.78

Best,
Lance

On Tue, Oct 25, 2016 at 4:46 PM, Mark Jenkinson
<[log in to unmask]> wrote:
> Hi,
>
> We don't really have a method that calculates TIV.  So comparing anything from FSL with TIV from some other package would not be very informative.  However, if you wanted to compare how various different measures performed as regressors in statistical correction or normalisation factors, then we would recommend using the VSCALING value from SIENAX.
>
> All the best,
>         Mark
>
>
>> On 23 Oct 2016, at 23:52, Lance Stevens <[log in to unmask]> wrote:
>>
>> Thank you Mark, especially for the detailed responses. I can see it
>> would be inappropriate to reintroduce TIV as a nuisance regressor in
>> the VBM or First vertex analyses.
>>
>> Initially, if I simply wanted to compare TIV as calculated in FSL, vs
>> TIV as calculated in Free Surfer (mri_segstats --subject subjid
>> --etiv-only), would SIENAX still be more appropriate than FAST?
>>
>> Thank you again.
>>
>> Lance
>>
>> On Sun, Oct 23, 2016 at 4:11 PM, Mark Jenkinson
>> <[log in to unmask]> wrote:
>>> Hi,
>>>
>>> See answers below.
>>>
>>>> I have 3 questions related to FAST and SIENAX
>>>>
>>>> Re FAST: While this link http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FAST/FAQ indicates that extracting a good estimate of CSF should involve use of a T2 (rather than T1) to obtain maximum contrast between skull and CSF, I assume an accurate sum of total intracranial volume (TIV) be calculated from a T1 by simply summing the 3 partial volume estimates of all three tissue type, CSF, GM and WM? That a T1 is sufficient is indicated by the FAST practical. I would expect that the TIV would include the meninges. Are both of these assumptions correct?
>>>
>>> TIV strictly should include everything inside the skull but not the bone.  However, you can't really distinguish bone from other dark air/tissue areas in a T1 clearly and so a T1 is poor for getting a precise TIV calculation.  The brain extraction that is done on T1 will exclude some of the CSF and other tissues inside the skull, and so it does not give a precise TIV.  T2 is better as the CSF is bright and so at least this can be clearly distinguished from bone. But the good news is that you normally don't need to have a precise TIV measurement to do things like correcting for head size, and so a reasonable surrogate, as you would get from a T1 is normally sufficient for this.
>>>
>>>> Re Sienax: Sienax is indicated for total brain tissue volume estimation in pathology, such as Alheimers. Is this referring to TIV or just the GM and WM (in some contexts CSF and also referred to as a tissue). The practical ( https://fsl.fmrib.ox.ac.uk/fslcourse/lectures/practicals/seg_struc/index.html#sienax ) suggests that Sienax takes TIV (CSF, GM, WM) but extracts just the GM/WM total combined volume. Is this correct?
>>>
>>> The main result from SIENAX is a brain tissue volume - which is just GM + WM.  It uses registration based on a skull estimate to correct for head size, but does not calculate TIV explicitly.  However, this surrogate for head size is sufficient to correct for the majority of the effects related to head size.
>>>
>>>> I have two groups: 10 patients with intermediate stage Parkinson’s Disease (PD) and 10 controls. I want to treat TIV as a nuisance covariate and remove is contribution to VBM cortical and FIRST subcortical estimates.  I would assume FAST could be applied to both PD and controls to arrive at TIV estimates that could be run in an ANCOVA model to eliminate contribution to the VBM and FIRST outcomes?
>>>
>>> We would recommend using the VSCALING output from SIENAX as the better surrogate for head size, rather than trying to use FAST to calculate TIV.  You can then use the VSCALING values in place of TIV values to perform an equivalent correction.  However, also note that VBM gives you values that are related to volumes in standard space and hence are already largely corrected for head size.  It is fine to put in a regressor in your analysis to remove anything left over (as correcting twice in this way will not be a problem) but do not divide the values by your VSCALING value (or TIV or any other surrogate) as that would reintroduce a head size dependence, which you do not want. So stick to linear corrections within the GLM (or equivalent regression models).  For FIRST the volumes are in native space, and so you can divide these or correct via regression, but the vertex analysis results can either be in native space or in standard space, depending on the options you've used.  If they are in standard (MNI) space then they are already corrected and it is like VBM - correcting again via regression is fine but normalising again by division would reintroduce a dependence and so should be avoided.
>>>
>>> All the best,
>>>        Mark
>>>
>>>
>>>
>