Scientific Lectures //
Accounting for some Prior Information when using Sequential Dictionary for fMRI Data Analysis
Karim Seghouane, Ph.D., ARC Future Fellow, Department of Electrical and Electronic Engineering, University of Melbourne
Presented: October 10, 2016
ABSTRACT: Sequential dictionary learning algorithms have been revealed as successful alternatives to conventional data driven methods such as independent component analysis (ICA) for functional magnetic resonance imaging (fMRI) data analysis. fMRI datasets are however structured data matrices where the structure can be characterized by the prior information on the fMRI data sets. This information has not been included in dictionary learning algorithms when applied to fMRI data analysis. In this seminar, I will present three dictionary learning algorithms dedicated to fMRI data analysis by accounting for this prior information. As most dictionary learning they are two stages algorithms and they differ from standard dictionary learning algorithms in both sparse coding and dictionary update stages. The first two algorithms account only for the known correlation structure in the fMRI data by using the squared Q, R-norm instead of the Frobenius norm for matrix approximation. The third and last algorithm account for both the known correlation structure in the fMRI data and the temporal smoothness. The temporal smoothness is incorporated in the dictionary update stage via regularization of the dictionary atoms obtained with penalization. The performance of the proposed dictionary learning algorithms are illustrated through simulations and applications on real fMRI data.