CMStatistics 2016: Start Registration
View Submission - CMStatistics
B0512
Title: Convergence of FastICA algorithms in performing temporal ICA for fMRI data Authors:  Jari Miettinen - University of Jyvaskyla (Finland) [presenting]
Klaus Nordhausen - University of Jyvaskyla (Finland)
Sara Taskinen - University of Jyvaskyla (Finland)
Klaudius Kalcher - Medical University of Vienna (Austria)
Roland Boubela - Medical University of Vienna (Austria)
Abstract: The most popular method in independent component analysis (ICA) is called FastICA, which has two classical versions with different approaches to find the independent components. In deflation-based FastICA the components are extracted one by one and in symmetric FastICA simultaneously. Both versions suffer from convergence problems when the sample size is small, which is one reason why temporal ICA has remained less popular than spatial ICA in the analysis of fMRI data. A small modification of the deflation-based algorithm, that is implemented in a function of R package fICA, improves the convergence remarkably. We have tested the convergence of FastICA algorithms in performing temporal ICA for 50 task fMRI data sets with 274 time points.