This thread raises two issues that are near and dear to my heart so I
thought I'd add my thoughts.
First, a conjunction analysis is not the same as using inclusive masking.
The former is a statistical test while the latter is a descriptive method.
In other words, given orthogonal contrasts as originally described, you
can ask the question what are the areas of common activation between the
contrasts of activation > baseline (A1>B1 and A2>B2). The formal
descriptions of this was worked out by Worsley and Friston (2000) and is
not implemented in FSL or MEdx. The output is a probability associated
with rejecting the null hypothesis (ie that they are activated together by
chance).
Inclusive masking, on the other hand, doesn't give you the probability
that you want -- instead it provides the probability of a significant
cluster/voxel based on a single contrast (A1>b1 if the mask was A2>B2 and
vice versa, otherwise). Because the threshold level of the masking is
arbitrary, this is not necessarily a rigorous test. Nonetheless, it can
be very useful in certain circumstances such as when the contrasts of
interest are not orthogonal. Hopefully both of these functionalities,
conjunctions and masking, will be added to FSL when the "visualization"
software becomes available (ie when we're not stuck using Medx to see
what FSL computed).
Joe
---
Joseph Devlin, Ph D.
FMRIB Centre, Univ. of Oxford
John Radcliffe Hospital
Headley Way, Headington
Oxford OX3 9DU
|