Meeting Banner
Abstract #1660

The Dependence of ICA Decomposition on Dimensionality in Functional Connectivity

Erik B. Beall1, Katherine A. Koenig1, Mark J. Lowe1

1Imaging Institute, Cleveland Clinic, Cleveland, OH, USA

The use of ICA in functional connectivity has been examined by others and a dependence on algorithm, algorithmic parameters and initialization/convergence has been shown, but the dependence on number of components has not been looked at beyond the consistency of dimensionality estimation methods. There are as yet no trusted methods for estimating dimensionality, so we demonstrate the effect of a range of dimensionality for four different commonly used algorithms in functional connectivity datasets and compare results with activation-seeded connectivity for motor, word generation and working memory tasks. Our results indicate a large effect that varies by function explored.