Thank you for the overview of the labeling rules. I have also tried setting WRITE_POSTERIORS to 1 to write out the posteriors.

I had a follow up question about obtaining the probabilities for the final subfield labels as available in the ?h.hippoSfLabels-T1.v10.mgz

I use mri_binarize on ?h.hippoSfLabels-T1.v10.mgz to obtain binary masks for each subfield, and then used these masks with their respective posteriors to obtain the distributions.

From: freesurfer-bounces@nmr.mgh.harvard.edu <freesurfer-bounces@nmr.mgh.harvard.edu> on behalf of Iglesias Gonzalez, Eugenio <e.iglesias@ucl.ac.uk>
Sent: Tuesday, March 21, 2017 2:22 PM
To: Freesurfer support list
Subject: Re: [Freesurfer] Hippocampal Subfields Posterior Probability

Dear Prad,

There is a “secret” way of getting the posteriors. You need to set the environment variable WRITE_POSTERIORS to 1 before you call recon-all.

In (t)csh:

setenv WRITE_POSTERIORS 1

In bash:

export WRITE_POSTERIORS=1

And then call recon-all -hippocampal-subfields-T1(T2) as usual.

The discrete segmentation picks, for each voxel, the most likely label. The posterior of this label might or might not be above 0.5. For instance, if a voxel has p=0.4 of being CA1, p=0.3 of being background, and p=0.3 of being fimbria, it will be labeled as CA1 (even though p<0.5 for such subfield). Some people seem to prefer a 2-stage approach, in which a hippocampal mask is first created by selecting the voxels for which p(background)<0.5, and then each voxel within that mask is colored with the most likely subfield at that location.

I hope this helps!

Cheers,

/Eugenio

Juan Eugenio Iglesias

ERC Senior Research Fellow
Translational Imaging Group
University College London
http://www.jeiglesias.com
http://cmictig.cs.ucl.ac.uk/

On 21 Mar 2017, at 20:09, Bharadwaj, Pradyumna - (prad) <prad@email.arizona.edu> wrote:

Hello Dr. Iglesias,

I'm currently examining the distributions of the posterior probabilities of the different hippocampal subfields, and I'm interested in obtaining the probability distribution images for the subfields that correspond to the volumes output in the subjects text file.

While I can use mri_binarize on ?h.hippoSfLabels-T1.v10.mgz to obtain the subfield masks and use them to obtain the posterior probabilities distribution after soft segmentation,

is there some other straighforward way to obtain these posterior probabilities for the volumes stated in the output text file ?

A follow up question - It appears that the soft segmentation approach approximates thresholding the posteriors by a value around 0.5 ? In your opinion, is this value acceptable and have you found similar values in your testing ?

Thanks!

-Prad

_______________________________________________
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.