Radu Mutihac
University of Bucharest, Romania
Title: Exploratory Analysis and Statistical Assessment of Functional Brain Imaging Data
Biography
Biography: Radu Mutihac
Abstract
Two approaches are employed in functional neuroimaging to reveal statistical regularities that can be associated with brain function: hypothesis-driven and data-driven analysis. Most functional neuroimaging data display significant higher-order statistics representing tendencies to grouping along various shapes, even if such feature is commonly hidden by the overall distribution. Exploratory data analysis (EDA) methods like independent component analysis or fuzzy cluster analysis reveal task-related, transiently task-related, and function-related activity without reference to any experimental protocol and including unanticipated/missed activations as well. EDA techniques can be regarded as hypothesis generating methods based on the underlying structure in data: a representative time course of activation for a set of voxels acts as an alternative hypothesis to the null hypothesis H0 (no activation). Any exploratory approach of neuroimaging data comes out with a set of estimated projections irrespective of the validity of the working hypothesis. As such, some means to statistically validate the data subspace is needed. Resampling is a statistical method producing surrogate data sets, which allows to approximate each parameter of the population by repeated learning of that parameter. Exploratory methods may reveal interesting features reporting on brain activations, whereas confirmatory methods are still necessary to validate the models and their statistical significance. Resampling based on prewhitening transform driven by an explicit noise model proves robust in the presence of BOLD signal. Repeated decompositions of functional neuroimaging data sets by different EDA algorithms reveal the stability/reliability of projections found.