Hey Doug, Bruce, et al., any update on why fscalc and Matlab are not producing the same values when trying to run a mean?  I’m using fscalc to pilot some stuff now, but not sure if I should be trusting the values I’m getting out of it.

It seems this issue might be caused by how Freeview is reading the images and not by either Matlab or fscalc.  See the screenshot below.  The two overlays loaded in Freeview should be the exact same image, which is the mean of three other images, one calculated with fscalc (*smMean246.mgh) and the other by reading the images into Matlab (*matlab_lh.mgh).  Notice that at the crosshair (vertex 37431) the values in the two overlays are different by ~0.03, or ~3.5%. 

However, if I read these two images into Matlab with MRIread() and look at that vertex, both images have the same value, which is 1.9042

So it seems like maybe Freeview and Matlab are not reporting the same values for the images, and possibly the images I write in Matlab are somehow getting depressed by Freeview.  I’m not able to stay connected to my cluster for much longer than 10-15 min at the moment, so it’s hard for me to troubleshoot more, but I can provide sample images and code to recreate the problem tomorrow.

Is this a known issue that somehow images written in Matlab are not read properly by Freeview?

This is on v5.3.0 and Matlab 2014b.  I can try tomorrow if the issue persists in v6.0

Thanks,

Jared

From: <freesurfer-bounces@nmr.mgh.harvard.edu> on behalf of "Zimmerman, Jared" <jaredz@pennmedicine.upenn.edu>
Reply-To: Freesurfer support list <freesurfer@nmr.mgh.harvard.edu>
Date: Friday, February 16, 2018 at 2:12 PM
To: Freesurfer support list <freesurfer@nmr.mgh.harvard.edu>
Subject: Re: [Freesurfer] Freesurfer equivalent to fslmaths?

Thanks Doug and Bruce, 

Fscalc is great because I can do maths with multiple inputs, like a mean.  I’m noticing, however, that when I do a mean with fscalc I get different values than when I do the mean in Matlab, any idea why?  In some regions the differences are as much as 5-10% so I don’t think it’s just a rounding/precision issue.

fscalc a.mgh add b.mgh add c.mgh add d.mgh add e.mgh div 5 --odt float --o mean.mgh

Hard to imagine it’s an order of operations issue, but I am a little confused about how fscalc handles order of operations from the help page.

I’m using v5.3.0 and Matlab 2014b

Thanks,

Jared

From: <freesurfer-bounces@nmr.mgh.harvard.edu> on behalf of Douglas Greve <greve@nmr.mgh.harvard.edu>
Reply-To: Freesurfer support list <freesurfer@nmr.mgh.harvard.edu>
Date: Friday, February 16, 2018 at 11:31 AM
To: "freesurfer@nmr.mgh.harvard.edu" <freesurfer@nmr.mgh.harvard.edu>
Subject: Re: [Freesurfer] Freesurfer equivalent to fslmaths?

Or fscalc

On 2/16/18 11:22 AM, Bruce Fischl wrote:

Hi Jared

I think mris_calc does at least some of what you want.

cheers
Bruce
On Fri, 16 Feb 2018, Zimmerman, Jared wrote:

Hi all,



Is there an equivalent of fslmaths in Freesurfer?  I would like to add two scalar value images (.mgh
files) that are registered to the fsaverage6 surface but I’m not seeing an obvious way to do it. 
Right now I’m reading the images into Matlab to add them, but this is a bit inconvenient because
what I would like to do is smooth an image by a small amount, add the original image back to it,
then smooth again marginally and iterate until I get to a target fwhm.  Since I can’t smooth inside
Matlab this necessitates writing out a temp image for each smoothing step then reading it back into
Matlab for the adding.  Obviously this is a solvable problem, but as someone only marginally
proficient in Matlab it’s something I’d like to avoid, plus it seems like a lot of I/O for this
task.



A little more detail on my data and what I’m trying to do:



The scalar images I’m working with are network confidence maps, basically like the spatial maps from
an ICA dual-regression.  I want to combine the confidence maps together into a hard partition and
write it to an annot file, but I want to smooth them first.  I’m concerned, however, that smoothing
is going to bias the parcellation against small network parcels and in favor of large network
parcels because in each confidence map the small parcels will be surrounded by lots of zeros (does
this make sense?).  To correct for this, my idea was to iteratively smooth by small amounts and to
add the original confidence values (or some fraction of them) back to the smoothed map after each
iteration so that regions of high confidence with a small/narrow spatial spread do not become
diluted by the smoothing and don’t get taken over by larger high confidence regions in nearby
networks.



One final question would be how to smooth on a surface without resampling.  Right now I’m using
mri_surf2surf and smoothing when I resample to the native mesh, but if I take the above approach I
would not want to resample at each smoothing step.  Could I just use mri_surf2surf with –srcsubject
and –trgsubject pointing to the same subject?





Thanks,

Jared

____________________________

Jared P. Zimmerman

jaredz@pennmedicine.upenn.edu

Neuroscience Graduate Student

Oathes Lab

University of Pennsylvania

_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer