Print

Print
Hello Stefano
 
I looked a bit at the segmentation with FAST using priors as well. I am not an experienced user, but the FSL package also contains a script called 'sienax' which makes things easier for starting users. Within this script registration to talaraich is first performed, followed by the segmentation. So using this script you won't suffer from misalignment of your image with the templates used for segmentation. Afterwards your segmented image is returned to the original position of your input image.
 
If you want to use a deskulled image for segmentation, you can do this as well. Deskull the image, perform sienax and afterwards put the skull back into the segmented image.
 
Note that if you use sienax, look for the line containing the command 'fast' and add '-a' or '-A' depending on how to use the priors (as initialisation = '-a' or also as a 'weight' for the final segmentation = '-A').
 
Probably other more experienced users can help you better, but I hope my comments can help you a bit further,
 
Regards
Ronald Boellaard

 

Clinical Physicist Nuclear Medicine/PET

Department of Nuclear Medicine and PET Research

VU University Medical Center

Amsterdam, The Netherlands

Phone: +31-20-4442887

Fax: +31-20-4443090

 
 
 
-----Original Message-----
From: Marenco, Stefano (NIH/NIMH) [mailto:[log in to unmask]]
Sent: Thursday, March 04, 2004 06:12 PM
To: [log in to unmask]
Subject: [FSL] FAST segmentation using priors

I have a question relating to segmentation with FAST.

 

I have been trying to use prior probabilities for initialization and for final segmentation in order to obtain more accurate segmentation of the basal ganglia, particularly the thalamus. I assume that in order to use prior probability masks the program tries to register the volume to be segmented to a template.

Can the segmentation go very wrong if the volume to be segmented is tilted by a large angle in respect to standard space? I think FAST works better on deskulled volumes, but the skull can provide useful info for accurate registration to a template, so if I want to use priors, should I deskull the volume before or after segmentation?

What happens if the brain I want to segment is much smaller or deformed in unusual ways as compared to the template? Can this result in inaccurate segmentation, due to poor registration with the template?

 

Thanks for your help....

 

Stefano Marenco, MD

Senior Staff Fellow

Clinical Brain Disorders Branch

NIMH

10 Center Drive, room 4S235

Bethesda, MD 20817

tel. (301) 435-8964

fax. (301) 480-7795

email: [log in to unmask]