SPM vs. DPARSF: order of steps and how do I remove nuisance covariates before extracting seed time courses (is code accessible)?

Dear DPARSF team,

I have analysed my resting state data with two different approaches:

1. using the DPARSF toolbox

2. using in house scripts for Matlab/SPM. 

Interestingly the results are fairly the same expect that the t-values on group level are much higher in DPARSF (FWE .05 k10, tmax= 104,67,
(seed 51, 12, -27)) than for SPM (FWE .05 k10,  tmax=63.08(seed 51, 12, -27)). 

The preprocessing steps were almost the same (not possible to have such huge effects). When I understood correctly in DPARSF the grey matter is corrected for nuisance covariates (global signal, WM, CSF, movement) BEFORE(!) the seed time course is extracted and the functional connectivity maps are build.

In SPM I included the nuisance covariates (WM, CSF, global signal, OOB, movement) in the sinlge subject GLM together with the "uncorrected" seed time course. Could this be the reason?

What could be the reason for the differences? How can I remove the covariates out of the grey matter signal and extract the time courses afterwards? Is there a script available (e.g. the way DPARSF is doing it)?

Help would be very much appreciated!

All the best, 




The most important difference is you used BETA values in SPM (if I guess correctly how you used multiple regression analysis).

However, in DPARSF, the value you got is a pearson' correlation coefficients.

The major differnece between beta and r, is if the variance of the two time series are normalized or not.