Best, Kersten

Am 04.03.2022 um 14:11 schrieb Julian Macoveanu <julian.macoveanu@gmail.com>:

External Email - Use Caution

(Apologies for reposting due to lack of answer)

Dear all,

This is my first attempt to analyze longitudinal structural data with
freesurfer, and I just do my best to follow the tutorial. I have some
things not clear, so any help would be appreciated.

We have a longitudinal study design with 2 groups, healthy and
patients who were scanned at baseline a second time following a
variable time interval (6 to 36 months). We want to investigate
whether the patient’s cortical thickness/volume progresses differently
compared to controls (group-by-time interaction effects). I chose to
analyze the data with LME due to variable follow-up time, having only
50% of the patients with a second scan, and also having family members
within and between groups.

Questions
1) To account for familial relationship, should I add a family ID as
covariate that stays in the design matrix X and count this covariate
as random effect when the model is fitted? Like this:
lhstats = lme_mass_fit_Rgw(X,[1 2 3],Y,ni,lhTh0,lhRgs,lhsphere);

Where the first 3 covariates in X are counted as random effects: 1 the
intercept, 2 follow-up time and 3 family ID.

The X would further include group, age, sex, TIV, and group*time, the
last one being the interaction term we want to assess using CM.C = [0
0 0 0 0 0 0 1].

Does the X and CM look good to address our research question?

This is a three-level model, right? Time within patients, and patients within families - correct me if I am wrong.

In general, including ‚family‘ as a random effect is plausible in my eyes.

However, I have to admit that I am not sure at the moment how the lme tools handle these higher-level (i.e. more than two levels) models, in particular if nested random effects (patients within families) as opposed to crossed random effects are present.

So if it is scientifically justifiable to leave out ‚family‘ as a random effect, you have a simpler, two-level model and are on the safe side. Otherwise, if you’d like to stick with the more complex model, I’d need to take a closer look, and it may not be supported by the toolbox.

Beyond that, I always like to also look at the most simple model, which just has ‚patient‘ as the single random effect (i.e. assuming parallel slopes within groups in your case). It may give more stable estimates, in particular for small samples.

Apart from these considerations about which effects to include, technically X and CM look OK to me.

2) It is unclear to me how to perform correction for multiple
comparisons and assess what is significant. I follow this procedure
after constructing X:

[lhTh0,lhRe] = lme_mass_fit_EMinit(X,[1 2 3],Y,ni,lhcortex,3);
[lhRgs,lhRgMeans] = lme_mass_RgGrow(lhsphere,lhRe,lhTh0,lhcortex,2,95);
lhstats = lme_mass_fit_Rgw(X,[1 2 3],Y,ni,lhTh0,lhRgs,lhsphere);
F_lhstats = lme_mass_F(lhstats,CM);
dvtx = lme_mass_FDR2(F_lhstats.pval,F_lhstats.sgn,lhcortex,0.05,0):

The dvtx is not empty but has 1 value. What I don’t understand, when
constructing the sig.mgh map it appears the FDR2 correction info is
not used:
fs_write_fstats(F_lhstats,mri,'sig.mgh','sig');

Yes, your intuition is correct, the ’sig.mgh’ file contains uncorrected p-values.

How can I be sure when I import sig.mgh in Freeview I only look at the
vertexes surviving multiple comparisons? The sig.mgh file shows a lot
of regions when thresholded at min 1.3

There are two ways for doing the multiple comparison correction: either transform the p-values from uncorrected to corrected, or just transform the threshold, and use the adjusted threshold on the uncorrected p-values.

The second approach is described in the tutorial; maybe take a look at the very end (i.e., the ‚pcor‘ threshold).

3) Given a few clusters showing an interaction effect which are saved
as sig.mgh and overlaid on surface in freeview, how do I know
(visualize) what is going on in these clusters, like which of the
group means go up or down over time? (I use FS71 and tksurfer not
working). Is it here that printing out and plotting the betas (of the
interaction term) are useful for?

Yes, that is one motivation for extracting the betas. An advantage is that the betas will also inform about the magnitude of the effect, so this is preferred way in my opinion.

As an alternative, the output from the ‚lme_mass_F‘ script, e.g. the ’F_lhstats’ structure, contains a field called ‚sgn‘, which is the sign of the contrast - in the simplest case just the sign of the beta. So this might be used as well.

4) I have a hypothesis to find differences in prefrontal cortex, at
which step would I apply a PFC mask? I have constructed a PFC.label
file by adding corresponding label files.

I would do that just prior to multiple comparison correction: You can use a mask for the ‚lme_mass_FDR2‘ script.

Thank you,
Julian

_______________________________________________
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 Mass General Brigham Compliance HelpLine at https://www.massgeneralbrigham.org/complianceline <https://www.massgeneralbrigham.org/complianceline> .
Please note that this e-mail is not secure (encrypted). If you do not wish to continue communication over unencrypted e-mail, please notify the sender of this message immediately. Continuing to send or respond to e-mail after receiving this message means you understand and accept this risk and wish to continue to communicate over unencrypted e-mail.

--
Kersten Diers
Image Analysis Group (AG Reuter)
German Center for Neurodegenerative Diseases (DZNE)
B.1.114 | Building 99 | Venusberg-Campus 1 | 53127 Bonn | Germany
MailScanner has detected a possible fraud attempt from "secure-web.cisco.com" claiming to be https://www.dzne.de/en/sites/bonn/research-groups/reuter.html
Phone: +49 / 228 / 43302 - 381