External Email - Use Caution
Hi Douglas,
Could you please reply my last email (sent to
freesurfer@nmr.mgh.harvard.edu
<mailto:freesurfer@nmr.mgh.harvard.edu>) regarding my question about
the weighted age? I am eager to listen to your answer/opinion to my
question. For your convenience, I also include the whole email chain
below this email.
Thank you very much!
Sincerely,
Xiao
<https://www.mail-archive.com/search?l=freesurfer@nmr.mgh.harvard.edu&q=from
:%22Douglas+N.+Greve%22> Douglas N. Greve
<https://www.mail-archive.com/search?l=freesurfer@nmr.mgh.harvard.edu&q=date
:20200507> Thu, 07 May 2020 14:56:08 -0700
sorry, please include the previous email chain so that I know what you are
referring to
On 5/7/2020 5:33 PM, Xiaojiang Yang wrote:
External Email - Use Caution
Douglas,
Just as I mentioned in the initial email, subjects in the healthy
group have
different ages. Because average cortical thickness is believed to be
declined as the age gets older, I want to "normalize" the group by
multiplying each subject's thickness with an age-dependent scalar
(pre-calculated constant). Suppose I take age 30 as the "standard"
age, then
subjects with ages greater than 30 will have scalars greater than 1
(depend
on age); subjects with ages smaller than 30 will have scalars less than 1.
In this way, all subjects in the group looks like to be at the same age
(30).
Of course, I also suspect that there is no need to do the above mentioned
normalization if I use mri_glmfit. But my initial intention was using
non-linear scalars, and mri_glmft can only use linear fit. Please give me
your opinion. Thank you!
Xiao
<https://www.mail-archive.com/search?l=freesurfer@nmr.mgh.harvard.edu&q=from
:%22Douglas+N.+Greve%22> Douglas N. Greve
<https://www.mail-archive.com/search?l=freesurfer@nmr.mgh.harvard.edu&q=date
:20200507> Thu, 07 May 2020 14:06:27 -0700
I don't know what weight is. Can you elaborate?
On 5/7/2020 4:54 PM, Xiaojiang Yang wrote:
External Email - Use Caution
Hi Douglas,
Thank you very much! Could you please help me with more detailed
information?
1)Should I use "Variables age" or "Variables age weight" in the fsgd file?
If I can use either of them, which one is better?
I don't know what weight is. Can you elaborate?
2)For the contrast file, the content "1 -1 0" you mentioned, it
corresponds
to the "Variables age" fsgd file, correct?
Yes
3)If I use the "Variables age weight" fsgd file, what will be the
content of
the contrast file? Is it "1 -1 0 0"?
Yes
Thank you again!
Xiao
<https://www.mail-archive.com/search?l=freesurfer@nmr.mgh.harvard.edu&q=from
:%22Douglas+N.+Greve%22> Douglas N. Greve
<https://www.mail-archive.com/search?l=freesurfer@nmr.mgh.harvard.edu&q=date
:20200507> Thu, 07 May 2020 11:55:27 -0700
I think it will work properly if you use the raw data as input and
create an
FSGD file with two classes: (1) the individual subject and (2) healthy
subjects. Also include age as a covariate. Use DOSS instead of DODS. Then
set your contrast to be 1 -1 0 and run mri_glmfit. I think that should
properly account for age.
On 5/7/2020 2:44 PM, Xiaojiang Yang wrote:
External Email - Use Caution
Dear FS experts,
I have a group of healthy subjects. Given a new individual subject, I want
to compare its cortical thickness to the healthy group so that I can find
where the thickness is abnormal (thickening or thinning).
I use fsaverage as the template subject to calculate mean and std of the
healthy group. Since subjects in the healthy group have different
ages, I do
a "normalization" process to all the subjects in the group BEFORE
calculating the mean and std. Thus the normalized healthy group can be
regarded as all subjects having the same "standard" age. The normalization
process is just multiplying some pre-obtained scale (varied by age,
but not
linearly ) to the file *?h.thickness.fwhm0.fsaverage.mgh* of each subject
that are already calculated by "recon-all --qcache" command. So the
mean and
std of the group are actually weighted in the sense of Freesurfer's
recon-all results.
My questions are:
1)If I do vertex-wise t-tests by simply comparing individual subject's
thickness to the group using mris_calc but WITHOUT using mri_glmfit, can I
still use mri_glmfit-sim to do multiple comparisons correction? If
yes, how?
2)If the answer is no for the above question, how should I use
mris_glmft to
implement my above thoughts so that I can use use mri_glmfit-sim?
Specifically, how can I use the weighted mean for the group?
3)Take a step back, if I consider the weight is linearly correlated
with the
age, how to design the fsdg and contrast file?
Thank you very much!
Xiao