Print

---

Hello list!

I am trying to convolve four sets of images with a "hand made" hemodynamic
response function (covariate A-C) using a simple regression analysis

I tried two approaches and got two different results and now I hope that
someone can explain the differences between the approaches to me

1) fMRI models, no conditions, one regressor (the "hand made" hr
function), applied high pass filter and furthermore I chose finite impulse
function but thought that will not influence the design matrix since I
only have a regressor and 0 conditions (*stupid me?*)

I used four sets

1a) 160 images covariate A

1b) 516 images  covariateB

1c) 66 images covariate C

1d) 66 images taken from 1b and covariate C
I got nice activations that fit my hypothesis in 1a-1d


Then I used the what I thought more straight forward way

2)basic models/simple regression

I used the four sets

2a) 160 images covariate A

2b) 516 images  covariateB

2c) 66 images covariate C

2d) 66 images taken from 1b and covariate C

I got similar activations in 2a-2c in comparison to 1a-1c but the t-values
for conditions 2a and 2b are much higher (ROI analysis).

I get much lower t-values for 2c than 1c...(again ROI) could this be a
result of the number of images? Or what?

And...I get totally different results for 2d than I got for 1d (no
significant voxel (2d) versus nice results in 1d)

So, could someone explain me the differences between the two approaches?

Which one would be the appropriate analysis?

By using the fMRI models/0conditions/one regressor approach...how can I
make sure that spm will only do a regression (image with covariate) and
nothing else?

The reason why I did the two approaches was that I can do a group analysis
with approach 1 ( x subjects and x covariates)...is there a way to do that
with approach 2?

Thanks for all your help in advance

Nadine