Dear Dr. Gang Chen,

Currently, I am analyzing visual-task fMRI data from a marmoset. It was a simple block design, ~16 sec task (visual stimuli) + ~16 sec resting block (in total of 512 time-points x 2s TR).

We performed the task on many different days (sessions) and in each day has many runs.

The simplest way (and the default way of AFNI?): we wanted to concatenate all sessions & runs and run a single GLM to examine the activation/deactivation.

However, concatenating all data results in a very big data that AFNI or our computer cannot handle appropriately. (There seems to be a max timepoint that AFNI can handle?)

Thus, we ran GLM on each session (only concatenating multi-runs within each session), and obtained t-statistics & coefficients for each session.

Now, what is the appropriate way to combine these multiple t-statistics & coefficients from different sessions?

1. Simply averaging the t-statistics & coefficients? (seems not statistically valid?)
2. run another 3dttest or 3dLME, etc on these t-statistics or coefficients? (If I understood correctly these programs will modeling the random effect as well, but our data is from one marmoset, and we should only examine the fix-effect?)
3. or other methods I don't aware?

Thanks, and look forwarding to your reply!



Cirong Liu