 |
 |
Hi Steve, thanks very much for your quick reply. We are still confused by a couple of things: - Our design involves 5 conditions, each presented twice for 30 seconds. All conditions are modelled (including crosshair). So the model is: CH A B C D A B C D CH. We did not use temporal derivatives. I hope this all makes sense. - All 5 PEs we specified in our design matrix (block design) are negative in our ROI. Thinking that our ROI may have a lower blood flow level than the brain overall as Joe Devlin mentioned, we ran avwstats on a couple filtered_func_img and get a mean of about 9900. But the raw data in our ROIs are typically around 11000-13000! Part of our study is repeating an earlier study of the hippocampus done with healthy subjects, analyzed using SPM. In this study, we get the expected results with FSL for new healthy subjects. We don't see the predicted group differences between patients and controls and we are trying to understand our data better. So given a task that seems to work in a complex contrast, we have trouble understanding how all PEs that were modeled are negative in value. Thanks, Dost On Tue, 26 Oct 2004 18:59:45 +0100, Stephen Smith <[log in to unmask]> wrote: >> - Doesn't the PE represent beta from the GLM? Is it possible that PEs >> would be negative because the blood flow changes are the opposite of >> what I modeled? > >That's all correct, yes. > >> - In my dataset, one group has consistently MORE negative PEs than the >> other (both have all negative PEs). Is this related to lower overall >> levels of blood flow in my ROI in one group, or to worse fit of BF >> changes to my model? > >It depends really on what, qualitatively, you have put into your model. >For example, if you had created an EV that was 1s during baseline and 0s >during activation then the interpretation would be simple.... negative PE >means standard positive activation..... So it doesn't necessarily follow >that your results mean lower overall levels of blood flow (it's going to >be relative anyway when considering a given EV) or a worse fit (worse >would normally imply closer to zero rather than more negative)....