Hi Jorge,
Following your instructions, so far we have done the following:
1-Read your label
lhcortex = fs_read_label('freesurfer/subjects/fsaverage/label/lh.cortex.label');
2-Read the data file
[lhY, lhmri] = fs_read_Y('lh.thickness.mgh');
3-Fit a vertex-wise lme model with two random effects for the intercept term
and time eg.:
lhstats1 = lme_mass_fit_vw(X, [1 2], lhY, ni, lhcortex);
4-Fit a vertex-wise lme model with two random effects for the intercept term
and cannabis use eg.:
lhstats2 = lme_mass_fit_vw(X, [1 3], lhY, ni, lhcortex);
lhstats3 = lme_mass_fit_vw(X, [1 2 6], lhY, ni, lhcortex);
%intercept_time_gender
lhstats4 = lme_mass_fit_vw(X, [1 2 7], lhY, ni, lhcortex);
%intercept_time_age
lhstats5 = lme_mass_fit_vw(X, [1 2 3 6 ], lhY, ni, lhcortex);
%intercept_time_cannabis_gender
We displayed the lREML data on the surface models in matlab. In some
cases,(when there were 3 or more effects ( i.e. 1 2 6) ) the lreml values had
real and imaginary values, so I displayed the ABS value of the lreml
We need to know the following:
1. How do we model this:
Intercept, time, age, gender, Alcohol, other drugs vs.
Intercept, time, age, Gender, Alcohol, Other drug, cannabis
2. Correct for multiple comparisons
3. Open these in Freesurfer, significance maps using tksurfer ( P < 0.05)
Is it only visual, or is there a significance test between the two models
4. How do we get a map that demonstrates the unique effect of cannabis
5. What Contrast matrix do we use for the LME_mass_F program
Thanks
Jon
________________________________
From: jorge luis <jbernal0...@yahoo.es>
Sent: Wednesday, December 17, 2014 9:25 AM
To: Freesurfer support list; Jon Alan Wieser
Cc: Krista Lisdahl Medina; alicia.thomas....@gmail.com
Subject: Re: [Freesurfer] longitudinal statistics LGI
Hi Jon
We recommend to order the columns of your design matrix in the following way:
First, the intercept term (which is a column of ones); second, the time
covariate; third, any time-varying covariates (eg. cannabis use); fourth, the
group covariates of interest (eg. a binary variable indicating whether the
subject is a patient or control) and their interactions with the time-varying
covariates; finally any other nuisance time-invariant covariate (eg. gender).
So your design matrix is comprised by the following columns:
1. Intercept (a column of ones)
2. Time since baseline
3. cannabis use (time-varying if varies over time for each subject during the
follow-up time)
4. alcohol use (time-varying if varies over time for each subject during the
follow-up time)
5. drug use over time (time-varying if varies over time for each subject during
the follow-up time)
6. gender
7. age at baseline
There is no GUI for setting up the models. Here is an outline of the basic
steps (with only three time points you shouldn't need more than two random
effects):
1-Read your label eg.:
lhcortex = fs_read_label('freesurfer/subjects/fsaverage/label/lh.cortex.label');
2-Read the data file eg.:
[lhY, lhmri] = fs_read_Y('lh.thickness.mgh');
3-Fit a vertex-wise lme model with two random effects for the intercept term
and time eg.:
lhstats1 = lme_mass_fit_vw(X, [1 2], lhY, ni, lhcortex);
4-Fit a vertex-wise lme model with two random effects for the intercept term
and cannabis use eg.:
lhstats2 = lme_mass_fit_vw(X, [1 3], lhY, ni, lhcortex);
And so on with other time-variying covariates...
Now see which model fit produces the best lreml values across vertices in
general and then:
4-Perform vertex-wise inferences using the winner model eg.:
CM.C = [your contrast matrix];
F_lhstats = lme_mass_F(lhstats_winner, CM);
5-Save results eg.:
fs_write_fstats(F_lhstats, lhmri,' sig.mgh', 'sig');
-Jorge
________________________________
De: Jon Alan Wieser <wie...@uwm.edu>
Para: jorge luis <jbernal0...@yahoo.es>; Freesurfer support list
<freesurfer@nmr.mgh.harvard.edu>
CC: Krista Lisdahl Medina <krista.med...@gmail.com>;
"alicia.thomas....@gmail.com" <alicia.thomas....@gmail.com>
Enviado: Martes 16 de diciembre de 2014 15:24
Asunto: Re: [Freesurfer] longitudinal statistics LGI
Jorge,
We are interested in examining the impact of cannabis exposure (time-varying
continuous variable) on local gyrification index over 3 time points (baseline,
18 month, 36 month)- so this is a time-varying random effect. I apologize in
advance if these are student questions… we are novices here…
From what you said previously, we would want to model intercept+time, vs
intercept+cannabis use, vs intercept+time+ cannabis use. Vs. intercept+time+
cannabis use.+covariates (alcohol use over time, gender, age, drug use over
time). We are trying to figure out how to do this in Freesurfer/Matlab using
the Wiki (https://surfer.nmr.mgh.harvard.edu/fswiki/LinearMixedEffectsModels).
There is a great example in there with AD/MCI groups, but none outlining an
example focused on time-varying continuous variables.
So with OUR question, would we organize the data as follows:
Intercept
Time
Cannabis total (changes over time)
Alcohol total (changes over time)
Other Drug total (changes over time)
Gender
Age at baseline
So, a few questions:
1) We need help writing the syntax for Matlab for testing the models
(assuming linear trend was significant):
a. intercept+time, vs
b. intercept+cannabis use, vs
c. intercept+time+ cannabis use. Vs.
d. intercept+time+ cannabis use.+covariates (alcohol use over time, gender,
age, drug use over time)…
i. For example, this is the
linear model for cortical thickness over time for the groups example: Yij = ß1
+ ß2*tij + ß3*t²ij + ß4*sMCIi + ß5*sMCIi*tij + ß6*sMCIi*t²ij + ß7*cMCIi +
ß8*cMCIi*tij + ß9*cMCIi*t²ij + ß10*ADi + ß11*ADi*tij + ß12*ADi*t²ij + ß13*E4i +
ß14*E4i*tij + ß15*Genderi + ß16*BslAgei + ß17*Educationi + b1i + b2i*tij+ eij
ii. This is the design
matrix: [lhTh0,lhRe] = lme_mass_fit_EMinit(X,[1 2],Y,ni,lhcortex,3);
2) Is there a GUI available for setting up these models? (We are assuming
there isn’t and that it is all matlab based.)
3) Once we test these models, is it correct that we open the spheres
representing the liklihood ratio test results (corrected for multiple
comparisons) and pick the “best” model based on the greatest #/size of
significant clusters?
Jon
Jon Wieser
Research Specialist
UW-Milwaukee
Psychology Department, Pearse Hall Rm 375
2441 East Hartford Ave
Milwaukee, WI 53211
Phone: 414-229-7145
Fax: 414-229-5219
________________________________
From: freesurfer-boun...@nmr.mgh.harvard.edu
<freesurfer-boun...@nmr.mgh.harvard.edu> on behalf of jorge luis
<jbernal0...@yahoo.es>
Sent: Tuesday, December 2, 2014 12:34 PM
To: Freesurfer support list
Subject: Re: [Freesurfer] longitudinal statistics LGI
Hi Jon
I guess that when you say “we have continuous data as to the amount of drug
usage” you actually mean that the amount-of-drug-usage is a continuous variable
that changes over time for each subject. So yes you can keep this variable as a
continuous variable. In fact it can even be a random effect in your statistical
model. You will need to select the model with the best combination of random
effects : intercept+time vs intercept+amount-of-drug-usage vs
intercept+time+amount-of-drug-usage.
Actually one nice feature of the LME model implemented in freesurfer vs
commonly used two-levels random effects models in neuroimaging is that you can
include this type of longitudinal continuous variables in the model for the
mean without requiring it be included in the model for the covariance (i.e
included as a random effect). You just select the best subset of random effects
as explained above.
-Jorge
________________________________
De: Jon Alan Wieser <wie...@uwm.edu>
Para: "freesurfer (freesurfer@nmr.mgh.harvard.edu)"
<freesurfer@nmr.mgh.harvard.edu>
Enviado: Martes 2 de diciembre de 2014 12:29
Asunto: [Freesurfer] longitudinal statistics LGI
HI freesurfer experts
I have a question about the statistical analysis of longitudinal data. we have
run our data through the longtudinal data processing stream.
We are looking at the longitudinal effect on the LGI data
We are looking at doing a Mixed effects analysis. our main Model Factor
(Independent Variable) of interest is drug usage. we have continuous data as
to the amount of drug usage. Can this variable be continous variable, or do we
have to break it up into discrete levels of usage ( example, low, middle,
high)
Thanks
Jon
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu<mailto:Freesurfer@nmr.mgh.harvard.edu>
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu<mailto:Freesurfer@nmr.mgh.harvard.edu>
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
