Dear FSL Users and Experts,
We are using melodic on block-design task data (tensor ICA), and looking at differences between two groups of about 40 subjects each. Because this is a clinical dataset, we've tried running and re-running the analysis with several subjects excluded and noticed that the component structure has shown more variability between these different subsets than we expected. Also, the components that are identified as being significantly different between the groups don't seem to be consistent between slightly different subsets of the data. Is there a convention for identifying outliers who might be driving these differences? In order to graph the data, we'd like to extract a subject level parameter for the components that are being identified as different between the groups. Does any such parameter exist?
Thanks,
Hollis
|