Dear all,
I am trying to compare condition A versus condition B in a resting-state fMRI study. There are 9 subjects, 5 runs each one.
The analysis has consisted of:
A) Preprocessing each run separately, without concatenation per subject
B) Group-ICA: concat approach
C) Dual regression, with design.mat and design.con set to 1, for group-mean (one-group t-test modelling)
D) FSLNets with dual regression stage 1 time series
D.1) Obtain netmats: L1-regularised partial correlation, one row per run
netmats = nets_netmats(ts,0,'icov',10) % No r to z transformation, as most likely will not generate true z-stats for regularized options
D.2) In nets_groupmean.m, average the runs of each subject in netmats (if there are 5 runs per subject, average 5 rows and obtain a single row per subject, output: netmats_subject)
D.3) In netmats_glm.m, paired t-test with netmats_subject and the design matrix attached
D.4) With nets_edgepics, search for pairs of edges (ICs) whose functional connectivity is significantly greater in one contrast (A > B or B > A), p < (0.05 / 2), since there are two contrasts.
I have two doubts, please could you help me to solve them?
1) Is right the analysis performed with dual regression (step C) and FSLNets (step D)?
2) Before group-ICA, would be adequate to do single-ICA for each subject (concatenating the corresponding 5 runs), removing non-interest ICs with fsl_regfilt and finally splitting runs?
