Cluster size as a measure of task similarity?

Dear Neurostars list,

I have fMRI data from an experiment with 3 conditions: A, B, C. I would like to define a measure of similarity for each pair of tasks. I am considering doing a Representational Similarity Analysis (RSA) later on, but for now I wonder if the following (rather simplistic) approach is valid:

  • compute cluster size (after corrections) for each contrast, at subject-level, using the same T-threshold and cluster extent
  • do a paired-samples t-test between the distributions corresponding to each condition in the contrast, to check for a difference between the means
  • if the t-test is significant, say in the direction (A-B) > (A-C), claim that A is “more similar” to C than it is to B (assume it’s ok for “similar” to remain incompletely specified at this stage).

This approach hinges of course on the assumption (which I am not 100% sure is correct) that the cognitive states/tasks corresponding to 2 conditions are all the more similar the smaller the cluster is for the contrast that comapres those conditions. So if the number of significant voxels in the “A-B” contrast is 1000, and only 500 for “B-C”, can I conclude that tasks B and C are more similar than A and B, in terms of the BOLD responses that they evoke?

Clearly, this type of simple comparison (number of significant voxels) says nothing about similarity/difference in the spatial patterns of activity, but I am unsure whether the number of significant voxels can even be taken as an index of the similarity/difference between two tasks.

Would very much appreciate any advice, thanks in advance!

Best wishes,