You might find it easier to use to model setup wizard with the perfusion rArA... option, this will give you the correct EVs. You can then modify EV2 with your custom shape if you need to (assuming that the default box car does not describe your task). As you have noted you will need to change the phase of EV1 to reflect that fact that you have control first rather than tag. You have not been able to do this with your EV1 as you have loaded a custom file - for perfusion you can model the tag-control sequence using a normal square shape with On and Off equal to the TR (and no convolution) - the wizard does this for you. The wizard will also setup the interation correctly, and sets 'Make Zero' for EV to Centre within EV3.
Michael C
On 29 Sep 2010, at 23:13, Natalie Paul wrote:
> Hello,
>
> I am collecting perfusion data during a task with 5 blocks. The task blocks are very long - about 40 seconds each. I am interested in analyzing the data with the full perfusion signal modeling option. My design starts with control first (not tag, which the tutorial assumes).
>
> Here is the tutorial I was using in setting up my design:
> http://www.fmrib.ox.ac.uk/fsl/feat5/perfusion.html
>
> I followed the instructions and set up 3 EVS: one modeling the onset of the control periods (EV1), one modeling the BOLD signal during the task block (EV2), and one modeling the interaction (EV3). EV1 has a custom shape, which was a text file I created with the onset time and duration of the control periods with a weighting of 1. I did not convolve this EV with any shape, because I thought it made sense to leave it as what it is - a box-car waveform. EV2 has a custom shape, which was a text file I created with the onset time and duration of my task blocks. I convolved this EV with a Double-Gamma HRF waveform. In EV3, I clicked "interaction" as the basic shape. From what I understand, an interaction is created by multiplying the design matrix from EV1 with the design matrix of EV2.
>
> Modeling the interaction. I first used a model setting the min of EV1 to 0 and the min of EV2 to zero, because I thought setting the min of both of them to zero would ensure the interaction only reflected the control periods during which the task would occur. I am calling this "model 1". However, I noticed that some of the PPI models have a similar set up to the perfusion data. In PPI analysis, one is looking for periods of activity that are occurring during both during the BOLD response and a psychological regressor. Using this thinking, I set the mean of EV1 to 0 and the centre of EV2 to 0. I am calling this "model 2".
>
> The fsl wiki on perfusion says several things that are confusing to me. Firstly, the wiki says if control is first (rather than tag), one should set the phase for EV1 to the TR. Given I am modeling the tag-control sequence without convolving it with a specific shape, there is no option to change the phase. The wiki does not mention convolving EV1 with any waveform, and the model on the site appears to show the authors used a boxcar waveform as well. Secondly, the wiki mentions clicking a "Demean EV" button, which I do not see anywhere in my FEAT5 GUI.
>
> Here are my questions:
> 1) How should I be modeling the interaction between EV1 and EV2? Should I be using model 1, model 2, or a different model entirely?
> 2) Should I be convolving EV1 with any shape?
> 3) How does having control first (rather than tag) affect the way I am modeling it if I am only putting the onset and duration of control periods in my custom EV shape?
>
> Thank you,
>
> Natalie Paul
>
Michael Chappell MEng DPhil MIET
Deputy Group Leader, PUMMA, Institute of Biomedical Engineering, University of Oxford.
http://www.ibme.ox.ac.uk/~webpumma T: +44 (0)1865 617657
Junior Research Fellow, St. John's College, Oxford.
http://www.sjc.ox.ac.uk
Associate, FMRIB Centre, University of Oxford.
http://www.fmrib.ox.ac.uk/~chappell T: +44 (0)1865 222523
NOISEmaker, New Outlooks in Science and Engineering
http://www.noisemakers.org.uk
|