 |
 |
Dear FSL experts, We have been working on a dataset containing 22 resting state fMRI scans (160 volumes per scan, 11 subjects were scanned under two conditions). We have a few questions: - When using melodic (temporal concatenation) FSL gives 138 components. From component 88 onwards, the percentages explained and total variances are negative. What does this mean? - When running separate melodic analyses for the two conditions (n=11 each), one condition gives 137 components while the other contains only 35 components. In both analyses negative explained and total variances can be seen. - Out of the analysis on all data, we chose a small number of the melodic components and extracted their time-courses. We used the time-course to reconstruct for each of the 22 scans separately the individual maps of the network depicted in that component using lower level feat. Those individual maps were then used as inputs in a higher level feat analysis to compare the two conditions in a ‘two groups paired design’. This was done with mixed effects stats and rerun with fixed effects stats. These different methods give very different results that do not even overlap. In melodic we already included a two groups paired model and made comparisons between the two conditions. The results from the higher level feat analyses do not seem to be consistent with the outputs generated by melodic. Is the method that we chose to use plausible? Is it possible to reconstruct the individual maps out of the melodic time-courses and then use these in higher-level analyses? Best and thanks, Ysbrand