I'm very much enjoying the ICA-AROMA tool. It has reliably improved my sensitivity in several activation and connectivity analyses. I am now trying to apply ICA-AROMA to simultaneous multi-slice (SMS) data that is much more rich than typical fMRI datasets, with voxel sizes of 2x2x2 mm and a TR of 1 sec (~10 min runs). With default ICA-AROMA settings, including automatic dimensionality estimation, I am typically getting >100 components. It is not uncommon for the algorithm to then recommend regressing out ~50-80 components, a considerably higher number than than reported by Pruim et al in developing the method. In applying ICA-AROMA to such rich SMS data, do you recommend keeping automatic dimensionality estimation as such or limiting ICA to a fewer number of components?
Thank you
|