Hi Bruce (and Jorge),
Yes, it's the wm surface. I have also done the analyses with the pial surface and the results are similar to wm surface.
To your second question: White matter volume increased over this time period (lme analysis; controls: logP = 8.49, patients: logP = 6.34).
Since the cortical analyses were done using lme, which can handle missing data, some of the subjects have only one time point. So I created a difference map for those subjects for whom we have data on both time points, to see if area on the first time point is consistently larger than on the second time point. Almost all subjects showed larger values on tp1 than tp2 and the maps of average area change (across subjects) confirm that. In addition, I ran an lme analysis with the same subjects and found results very similar to those for the entire sample.
Would you agree that this apparent reduction in cortical area seems plausible? There is a reduction over time in raw data, and pial surface area show the same trend as wm surface, and the lme analysis with only subjects that have data on both time points shows very similar results as the lme with all subjects. On the other hand, I suppose we wouldn't expect increased wm volume together with reduced area?
As for the effect size maps, I have worked on finding a way to represent change in area over time that is intuitive for a reader not familiar with FreeSurfer: I figured one solution could be to log transform the dependent variable (wm or pial area). This way the significance tests are done with log transformed data and for purposes of illustration I do exp(beta)*100-100 on the beta for time, which ensures that if there is e.g. a 1% reduction, the figure shows -1, and 1 for a 1% increase. I find this is a good way of demonstrating the effects (attached figure: lh_wmarea_logtransf_expBeta2_s30_inflated_lateral.tif ). What do you think?
I could of course also transform the dependent variable into percentages. That is, baseline == 100 and tp2 expressed in percent of baseline. However, I find this to be a less attractive solution because we basically lose the baseline values, and this makes the model less useful for all other purposes. For instance, we can't investigate group differences at the various time points within the model. Perhaps more importantly, it's unclear what assumptions we are making. The lme assumes a normal distribution and it's unclear to me what the distribution of such ratios are.
Thank you!
LMR
yours,
Lars M. Rimol, PhD Norwegian University of Science and Technology (NTNU) Trondheim, Norway
Bruce Fischl http://www.mail-archive.com/search?l=freesurfer@nmr.mgh.harvard.edu&q=from:%22Bruce+Fischl%22 Sat, 06 Sep 2014 07:00:14 -0700 http://www.mail-archive.com/search?l=freesurfer@nmr.mgh.harvard.edu&q=date:20140906
Hi Lars
which surface are you using? If it's the white surface you might try looking at white matter volume to see if it is decreasing
cheers Bruce
On Sat, 6 Sep 2014, Lars M. Rimol wrote:
Hi,
I have performed a longitudinal analysis using the lme module in FreeSurfer, with this model:
intercept(random effect) + centered age + group + group x centered age + sex
I tested the effect of time with this contrast vector [ 0 1 0 0 0 ]. Dependent variable is area.
Here, mapping the second beta means mapping the effect size for (change over) time. In the beta map, I find values from 0 to 0.004. I would interpret that to mean that local area shrinks by at most 0.004 mm² per year in the reference group. But I'm not 100% sure about the biological (or geometrical) meaning of that.
Can I interpret this literally as the mean yearly shrinkage of the three triangles surrounding a given vertex, the average of whose area comprises the area score of the vertex, being 4/1000 mm? Of course, these maps are smoothed with 30mm, so the real spatial resolution is nowhere near this....
Thank you!
-- yours, Lars M. Rimol, PhD St. Olavs Hospital Trondheim, Norway
Freesurfer mailing listFreesurfer@nmr.mgh.harvard.eduhttps://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 athttp://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.
yours,
Lars M. Rimol, PhD Norwegian University of Science and Technology (NTNU) Trondheim, Norway
On Mon, Oct 27, 2014 at 9:59 AM, Lars M. Rimol larilin@gmail.com wrote:
Hi Bruce (and Jorge),
Yes, it's the wm surface. I have also done the analyses with the pial surface and the results are similar to wm surface (attached p-maps: lh_0-1000_wmarea_s30_log10p_inflated_lateral.tif vs. lh_0-1000_wmarea_s30_log10p_inflated_lateral.tif).
To your second question: White matter volume increased over this time period (lme analysis; controls: logP = 8.49, patients: logP = 6.34).
Since the cortical analyses were done using lme, which can handle missing data, some of the subjects have only one time point. So I created a difference map for those subjects for whom we have data on both time points, to see if area on the first time point is consistently larger than on the second time point. Almost all subjects showed larger values on tp1 than tp2 and the maps of average area change (across subjects) confirm that (attached file: lh_diff_vo-v2_rawdata_lateral.tif shows timepoint1 - timepoint2). In addition, I ran an lme analysis with the same subjects and found results very similar to those for the entire sample (attached file: lh_2tp_01000_pmap_lateral.tif ).
Would you agree that this apparent reduction in cortical area seems plausible? There is a reduction over time in raw data, and pial surface area show the same trend as wm surface, and the lme analysis with only subjects that have data on both time points shows very similar results as the lme with all subjects. On the other hand, I suppose we wouldn't expect increased wm volume together with reduced area?
As for the effect size maps, I have worked on finding a way to represent change in area over time that is intuitive for a reader not familiar with FreeSurfer: I figured one solution could be to log transform the dependent variable (wm or pial area). This way the significance tests are done with log transformed data and for purposes of illustration I do exp(beta)*100-100 on the beta for time, which ensures that if there is e.g. a 1% reduction, the figure shows -1, and 1 for a 1% increase. I find this is a good way of demonstrating the effects (attached figure: lh_wmarea_logtransf_expBeta2_s30_inflated_lateral.tif ). What do you think?
I could of course also transform the dependent variable into percentages. That is, baseline == 100 and tp2 expressed in percent of baseline. However, I find this to be a less attractive solution because we basically lose the baseline values, and this makes the model less useful for all other purposes. For instance, we can't investigate group differences at the various time points within the model. Perhaps more importantly, it's unclear what assumptions we are making. The lme assumes a normal distribution and it's unclear to me what the distribution of such ratios are.
Thank you!
LMR
Douglas N Greve Tue, 09 Sep 2014 08:24:37 -0700
This is tough to interpret, but basically, yes it would be 0-.004mm2/year. It is not quite right to say that it is at that vertex because of smoothing, but in that area. It is also hard to say what the total change would be for a cluster. One could sum the changes over the cluster vertices, but that would probably over-estimate the change. doug
Hi Lars
which surface are you using? If it's the white surface you might try looking at white matter volume to see if it is decreasing
cheers Bruce
On Sat, 6 Sep 2014, Lars M. Rimol wrote:
Hi, I have performed a longitudinal analysis using the lme module inFreeSurfer, with this model:
intercept(random effect) + centered age + group + group x centeredage + sex
I tested the effect of time with this contrast vector [ 0 1 0 0 0 ]. Dependent variable is area. Here, mapping the second beta means mapping the effect size for(change over) time. In the beta map, I find values from 0 to 0.004. I would interpret that to mean that local area shrinks by at most 0.004 mm² per year in the reference group. But I'm not 100% sure about the biological (or geometrical) meaning of that.
Can I interpret this literally as the mean yearly shrinkage of thethree triangles surrounding a given vertex, the average of whose area comprises the area score of the vertex, being 4/1000 mm? Of course, these maps are smoothed with 30mm, so the real spatial resolution is nowhere near this....
Thank you! -- yours, Lars M. Rimol, PhD St. Olavs Hospital Trondheim, Norway
yours,
Lars M. Rimol, PhD Norwegian University of Science and Technology (NTNU) Trondheim, Norway
