 |
 |
leann helka kinnunen wrote: > Dear SPM Users and Authors: > I still have not heard back from anyone regarding my message from March > 30, so I thought I would reiterate my problem and ask if anyone else has > ever tried viewing the images which SPM performs its statistics on and > whether or not they found the same random patterns of zeros as I did. > I tried for the first time recently to use the "generate normalized > image data" option in SPM which should allow me to generate > and save the images resulting from the application of the global > normalization, scaling and gray matter threshold (in the statistics > portion > of SPM). I used a multi-subject, different conditions (PET) design type > with 6 subjects and 2 conditions each (a drug and a placebo). I used > proportional scaling with a gray matter threshold of 0.8 and an > uncorrected F threshold of 0.99. > Upon viewing the resulting PET images, I discovered that they showed a > random pattern of voxels throughout the brain with values of zero, having > no relation to white or gray matter areas. These patterns were the same > for every one of the images generated (even across subjects). The > realigned, normalized and smoothed (at FWHM of 12x12x12) PET images I > entered into my design matrix had no such pattern/problem. Once again I > ask if anyone has observed such a phenomenon before and whether anyone > would know the cause. I am concerned as I am not sure why such patterns > would be generated and how SPM statistics > would handle these random zeros when performing its analyses. > I would be more than happy to ftp my pre and/or post statistical analysis > images to anyone who would be willing to take a look at them. > > Thank you in advance for your help, > Leann Kinnunen > Department of Psychology > University of Chicago Hi Leann, It sounds like you are using MEDx. The normalized images correspond to the data in the XA.mat file generated during statistical analysis in matlab SPM. The data in these image are the result of removing confounds, global normalization, and scaling. Further, they are thresholded by the Gray Matter Threshold and masked by the F-map after it has been thresholded based on the UFp cutoff. I think you will find that if you lower the gray matter threshold to about 0.10 and leave the UFp threshold at 0.99 that these images will look more like the PET images that you started with. Also, you could use these images in the Eigen image analysis and you should find that the first or second components match fairly closely the major result from your statistical analysis. Best Regards, -John