Computationally Efficient Estimation and Inference Methods for Hierarchical ICA of fMRI Data

• Activity Number: 431
• Type: Invited
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: SSC
• Author(s): Ying Guo*+ and Ran Shi
• Companies: Emory University and Emory University
• Keywords: independent component analysis ; imaging ; EM ; computationally efficient estimation and inference ; fMRI ; subspace

Abstract

Independent component analysis (ICA) has become an important tool for identifying brain functional networks in neuroimaging studies. Recently, we have proposed a hierarchical group ICA model that provides a formal statistical framework for estimating and testing covariate effects in group ICA. A maximum likelihood estimation method via EM algorithm is proposed. However, our method faces a common challenge in imaging statistics, that is a heavy computational burden involved in statistical estimation and inference. In this talk, we present computationally efficient methods that can help address this issue. Specifically, we propose a subspace-based approximate EM, which can significantly reduce computational time while retaining high accuracy in estimation. Theoretical justification is provided for the proposed subspace approximation. Furthermore, to test covariate effects in ICA, we develop a voxel-wise approximate inference procedure which eliminates the need of computationally expensive covariance estimation. We illustrate the proposed methods via simulation studies and a real fMRI data example.