help > gPPI deconvolution work-around for event-related data
Sep 19, 2020  01:09 PM | Jessie Yaros
gPPI deconvolution work-around for event-related data
Hi there! I'm new to CONN and absolutely love it so far, but I've hit a bit of a roadblock and would love some advice.

I am planning on running whole-brain ROI to ROI gPPI analysis on a somewhat fast event-related design (My TR = 1.5s, with stimulus presentations every 4.5 seconds.) According to the CONN website, the toolbox uses the FSL ppi approach which does not deconvolve the BOLD timeseries before creating PPI interaction regressors. I have read that this is ok for block designs, but doesn't work so well on event designs (Gitelman et al 2003, O'Reilly et al 2012, Di et al 2017).

This leads me to the following questions:
A. Is there any way to create PPI interaction terms using the SPM decovolution method in Conn ?
B. If not, is there a way to recreate the ppi terms outside of CONN and somehow import and overwrite those for use instead of the default ones for the first level ppi GLM in CONN?
C.If this isn't possible, is the most reasonable approach to export denoised nii files out of CONN and run the full 1st level gPPI analysis in SPM and gPPI toolbox instead?
D. If the approach in option C is necessary, is there a good way to the first-level analysis results from SPM/gPPI toolbox  back into CONN for 2nd level analysis?

In addition could you clarify another related question for me about modeling of main task effects and its implementation in gppi?
E. It is my understanding that main of effects of task are always modeled in  gPPI analysis. However, it also recommended in CONN to model the main effects of task in the denoising pipeline. Therefore, does PPI analysis set the PRE-denoised timseries as the independent variable in the GLM, and include the main effects of task (calculated during the denoising step) as covariates? I just want to double-check that denoising is strictly used to create covariates for subsequent 1st level analysis, and not denoised timeseries that would be used as seed variables.I'm pretty certain the former is the case  based on these prior forum posts, but just want to be sure. 
https://www.nitrc.org/forum/message.php?...
https://www.nitrc.org/forum/message.php?...

Thank you in advance so much. I'd really like to be able to use CONN for second level analysis AND its graph theory applications. So i'm really hoping to find a work around! 

Best, 

Jessie

Threaded View

TitleAuthorDate
gPPI deconvolution work-around for event-related data
Jessie Yaros Sep 19, 2020
Jessie Yaros Sep 28, 2020