Unsupervised metrics of brain region significance for event-related fMRI intersession experiments

The non-invasive nature of Functional Magnetic Resonance Imaging (fMRI) has encouraged a large number of exploratory research studies that aim to identify regions of the brain that are involved in the workings of specific tasks. Conventionally, this kind of studies make use of supervised encoding methodologies, such as the General Linear Model (GLM), in which the contribution of different brain regions to a given task is studied as a function of the linear regression or correlation of the BOLD signal and the task regressors. Recently, decoding methodologies are taking the lead, as they allow for the use of unsupervised non-parametric approaches for the analysis of group fMRI datasets, such as Independent Component Analysis (ICA). A long standing problem with ICA techniques is the evaluation of the significance of the resulting spatial components that are involved in the underlying tasks that the subjects were performing in the scanner. In this paper, we describe the use of two different statistical association metrics for identifying significant components that result from a group ICA of event-related fMRI data. The suggested metrics have been evaluated against a real fMRI dataset in order to illustrate further their merits and drawbacks.