Dear FSL users,
I'd like to investigate voxel-wise differences in the local grey matter volume between two groups (patients vs controls) using VBM.
However, subjects come from 4 different sites ; each site has its own scanning protocol (in particular, voxel size and magnetic field strengths differ across sites).
Multi-site VBM feasibility seem to be discussed : some studies highlighted its feasibility, as long as I include site as co-variable in my statistical model, whereas others recommend prudence (see for example https://www.ncbi.nlm.nih.gov/pubmed/21324367).
First : from your experience, what do you think about multi-site FSL VBM feasibility ?
Second : if I decide to perform such analysis, I need some advices to create my study-specific template.
FSL-VBM user guide recommends : “if you have more than one group and the number of subjects in each is not the same, choose (at random) among the biggest group(s) the images that you will use to create the study-specific template, with the same number as of the smallest group (in order to create an unbiased template - see below for further explanation). Once you've chosen which T1 images to keep to build the template, put all the selected names of exams in a file called template_list in your FSL-VBM directory”.
Am I supposed to choose the same number of patients, controls in each site and then to create a template from these subjects ?
Let's imagine my database is as follows :
Site 1 : patients = a, controls = b
Site 2 : patients = c, controls = d
Site 3 : patients = e, controls = f
Site 4 : patients = g, controls = h
(When “a,b.,..,h”are numbers of subjects in each subgroup)
If d<a<f<b<c<g<e<h, am I supposed to choose “d” subjects in each subgroup (patients and controls) of each site to create my template ?
Thank you in advance for your help.
Best,
Franz
|