 |
 |
Dear Eraldo, >I am writing to seek advise on how to perform a seed based >correlation analysis on a group of subjects >scanned during resting state using SPM8 > >My intuition on how to proceed is the following > >1. to perform a box-car analysis at first level on every subject >defining, for example, the alternation of 2 conditions every 10 scans >2. to extract the BOLD signal fluctuations as a VOI of choice from >this analysis, the seed of interest >3. to perform a linear regression analysis between the values >extracted with the VOI and the rest of the brain >4. to bring the individual con-images generated by the linear >regression analyses to a second level analysis for group inference > about a shared correlation >5. of course, if one had two groups of subjects, a between group >comparison should also be possible > >I am a bit concerned on the validity of what I described at >point #1, the first level analysis That is absolutely right and properly infers on the subject-specific regression coefficients (not correlation coefficients) . Things get more interesting when you compare two groups like this: This is effectively a psychophysiological interaction (PPI; where the physiological variable is the activity in the seed VOI and the psychological variable is group) but implemented using the (two-stage) summary statistic procedure. I would not worry about step #1. Strictly speaking, you just need to coerce SPM into modelling the constant and drift terms only. We usually do this by specifying a single event (in an event-related design) and making sure the event is at the end of the session or run (before it is expressed hemodynamically). Do not forget to suppress the high-pass filtering if you want to preserve drifts or ultra slow fluctuations. Happy Christmas :) - Karl