 |
 |
Hello, Thank you for the quick reply! I don't know if I did something wrong because I haven't used Melodic earlier, but I did as follows: 1. First, I run FEAT again with "MELODIC ICA data exploration" box checked in the "Pre-stats" tab --> didn't change activation much, the same problem still exists! 2. Then I tried to run MELODIC ICA as on its own, not as a part of the FEAT. Thus I enclosed previously in FEAT run design.mat and design.con files to the MELODIC ICA (Post-stats tab), and of course typed Registrations/Pre-stat values. This gave me good-looking activations. The only problem is that, as is ICA's purpose, it calculates those components (504 components in my case) and they look good, but how am I supposed to know which contrasts (contrasts based on EVs I have created in FEAT with Custom 3 column format ) those components represent? The results page prints Contrasts and their p-value on each component's result page, but do I have to check each p-value of each contrast of the 504 components, or would it be possible to check which contrast are statistically significant and which are not? I have read the material in the Support and Course -sites but for some reason didn't find any answer to the question. Best regards, -----Original Message----- From: FSL - FMRIB's Software Library on behalf of Christian F. Beckmann Sent: Fri 3/28/2008 10:52 To: [log in to unmask] Subject: Re: [FSL] Problems with activation images Hi This looks to me like a scanner-related artefact - you could try running melodic on the data to see if it picks up on the problem. hth Christian On 28 Mar 2008, at 08:26, Markus Kivikangas wrote: > Hi, > > I'm a bit confused here. I run FEAT for several subjects and got > enclosed > post-stats (just two examples attached). Some of the subjects' > activations > are unrealistic (Screenshot 2), some are "normal" (Screenshot). > Moreover, > some subjects may have this unrealistic activation with e.g. > contrast 15 > another with e.g. contrast 14. I have used 14 EVs (Custom 3 column > format) > and 22 contrasts. Thresholding has been uncorrected P 0.05. What could > explain the problem? > <Screenshot1.jpg><Screenshot2.jpg>