An Evaluation of Independent Component Analyses with an Application to Resting State fMRI

• Activity Number: 362
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Benjamin Risk*+ and David Matteson and David Ruppert and Ani Eloyan and Brian Caffo
• Companies: and Cornell University and Cornell University and Johns Hopkins Bloomberg School of Public Health and Johns Hopkins University
• Keywords: Group ICA ; Nonconvex optimization ; Resting-state fMRI ; Stability analysis ; Stationary points

Abstract

We examine differences between independent component analyses (ICAs) arising from different assumptions, measures of dependence, and starting points of the algorithms. ICA is a popular method in functional magnetic resonance imaging (fMRI) studies. However, there are many ICA methods, and their solutions may differ. Infomax and FastICA are commonly applied to fMRI studies. Hastie and Tibshirani (2003) found that ProDenICA outperformed FastICA in simulations with two components. We introduce the application of ProDenICA to simulations with more components and to fMRI data. ProDenICA was more accurate in simulations, and we identified substantial differences between biologically meaningful ICs from ProDenICA versus other methods in the fMRI analysis. ICA methods require nonconvex optimization, yet current practices do not recognize the importance of, nor adequately address, sensitivity to initial values. We found local optima led to dramatically different estimates in both simulations and group ICA of fMRI. We applied the Hungarian method to match ICs from multiple estimates, thereby gaining novel insights into the impact of methodology and initial values.