Population inference: Create a sample for further analysis¶
Signal models have been fitted to FMRI session data:
fmrifit -v --stimulus-block letter --window 2 1 1
fsl4prune -v
The ATI reference field¶
Model based estimation encourages to analyse and report BOLD effects in above template image (ATI) units. The reason is that FMRI have no »natural« unit. If \((α_j(x), x∈M)\) is the intercept and \((β_j(x), x∈M)\) is the estimated task related BOLD effect in an FMRI, and if \((μ_x, x∈M)\) is an ATI reference field in standard space \(M\), then
is the BOLD effect at \(x\) in ATI.
As shown here, the statistical analysis of the task related BOLD effects benefits (by gaining power without loosing statistical validity) when the analysis is performed with the fields \(c_jβ_j\) and not with the fields \(β_j\). This is in particular true for multicentre studies.
A reasonable choice for the ATI reference field is to chose a field with similar curvature as the estimated statistics fields in the sample. As this curvature is predominately influenced by the choice of the spatial weighting scheme, a canonical choice is to smooth a template in standard space by the same spatial weighting that has been used when fitting the individual statistic fields.
Look for the parameter scale
in one of your fits:
fmriinfo results/WGT/AGB300/0002/AGB300-0002-WGT-2014-04-24-mni152-ants-pcm.fit
[...]
Hyperparameters:
Scale type: max
Factor: 3.00
Resulting in:
Mass: 0.9973
Scale: 2.30 mm
FWHM: 5.42 mm
Radius: 6.90 mm
Diagonal: 13.80 mm
[...]
Then create the corresponding ATI reference field:
nii2image --name mni152 \
/usr/share/data/fsl-mni152-templates/MNI152_T1_2mm_brain.nii.gz \
vb-template.image
fmriati 2.3 vb-template.image vb-ati.image -v
Create the sample for your study¶
fmrisample -v tutorial.sample --vb-ati-image vb-ati.image