Semi-blind FastICA of fMRI using temporal constraints

Abstract

Independent component analysis (ICA) is a data-driven approach that has been widely applied to functional magnetic resonance imaging (fMRI) data analysis. As an exploratory technique, traditional ICA does not require any prior information about the sources and the mixing matrix. However, it has been demonstrated that incorporating paradigm information into the ICA analysis can improve the performance of traditional ICA. In 2005, Calhoun proposed semi-blind ICA which improved the robustness of Infomax ICA in the presence of noises by regulating the estimated time courses with paradigm information. Different from the Infomax ICA algorithm, FastICA is able to estimating independent components one by one. If the target component can be estimated earlier, the FastICA algorithm can be terminated beforehand. Therefore, the order of the target component is important for FastICA to reduce computational time during one-to-one hierarchical estimation. In this paper, we proposed semi-blind FastICA by adding regularization of the first estimated time course using the paradigm information to the FastICA algorithm. We demonstrated the feasibility and effectiveness of our approach in extracting the task-related component from single-task fMRI datasets of block design. Results of both simulated and real fMRI data suggest that (1) In contrast to FastICA, the time of extracting the target component by semi-blind FastICA is largely reduced;(2) Semi-blind FastICA can accurately extract the task-related IC as the first one; (3) Semi-blind FastICA can estimate more accurate time course of the task-related component than FastICA.