help >

**MVPA questions more**Showing 1-1 of 1 posts

**190**SubscribersSep 14, 2012 06:09 PM | MVPA questions more Dear Alfornso, I still cannot get clear understanding of how conn performed PCA. Please, correct me if I am wrong (MVPA performed on three conditions, slow even-related design with random ISI, each condition is treated as a block of duration of 150 sec with 20 event onsets) 1. for each condition Conn takes BOLD response for each onset and convolves it with HRF (taking the only pic part of the HRF). 2.Correlates time-series for each voxel with the rest of the brain (result - a number of voxel-to voxel correlation-matrixes. The number of these matrixes is equal to number of scans) 3. Performed PCA (in form of SVD) for each voxel on the variability of connectivity with the rest of the brain (say we have 600 scans, then the PCA is performed on 600xnumber-of-voxels matrix). Here something is not clear, and I cannot get it from conn-process.mat: how the PCA scores are stored? We have one principal component for each voxel - how the information are stored in .nii file? It is not a time series, it is not a correlation. So, what is it? Understanding this step, I believe, will help to understand what information is contained in single subject/condition .nii. 4. On the second level we perform ANOVA /MANOVA(?) to get the group connectivity pattern (?). 5. Interpretation of the second level. Say, we have got a distinct pattern of voxel-to-voxel connectivity for each of our three conditions. In fact, each pattern tells us about correlated (synchronized) activity between all voxels comprises this pattern (right?)This pattern may consist of red and blue blobs. red blobs (positive correlation) tells us that the stronger BOLD response is associated with higher scores of the component, the blue blobs tell us that the weak response is associated with the lowest scores of the component. Am I wrong? Please, accept my apology if all above is wrong. But I cannot get this information from conn manual or matlab scripts. thank you in advance |