Hi David,
at TE 30ms you will have significant BOLD signal which will be picked up by Melodic.
W/r to your modelling, I would stick to full perfusion modelling.
Cheers,
Andreas
________________________________________
Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag von David Clewett [[log in to unmask]]
Gesendet: Donnerstag, 23. Juni 2011 18:30
An: [log in to unmask]
Betreff: [FSL] ASL resting state analysis
Dear FSL Experts,
I'm working with a resting state pASL dataset and have been analyzing functional connectivity using the GLM, seed-based FC, and MELODIC followed by dual regression methods. For 17 subjects, we've collected two resting state scans but in separate sessions: one with the experimental condition and one for the control condition; so basically 2 sessions per subject. The lower-level copes were then carried to a higher-level mixed-effects analysis (w/ paired t-tests).
I've attempted running multi-session temp. concat MELODIC with the "perfusion subtraction" option turned on or off. I've found that when the option is turned OFF, I acquire: more components, more network-looking components, and more significant components. Is this still a valid approach for pASL resting state data? I understand there is an issue with BOLD contamination in ASL, but I'm curious if our TE, which is 30ms, is low enough to mitigate this effect. Is it possible that the significant results are being driven by undesirable BOLD signal?
In my GLM, I've chosen to model the tag and control images as separate square waveforms within each session. By increasing the phase of "tag" - which is the first image type - by one TR, I've staggered the images so that they can be subtracted "manually" in a contrast. Essentially, the model is akin to the "c-t" column in the "full perfusion modeling" tutorial. This enabled me to carry lower-level feat contrasts to higher analysis (paired t-test between the two sessions: one stressed rest session and one control rest session). Does this approach necessitate the modeling of a separate BOLD EV? I've produced significant results, but would like to ensure that the activations aren't driven by BOLD contamination. I've also read that high-pass filtering can help remove these effects, but I'm still a little confused what cutoff is optimal (I'm currently using 100s, which is fairly typical for fMRI and mentioned on the website).
Sorry for such a long question, but any help would be very much appreciated!
Many Thanks!
David
|