Generalized mutual information fMRI analysis: a study of the Tsallis q parameter

The main objective of analyzing functional magnetic resonance imaging (fMRI) data sets is to screen the physiologically induced signals from noise or artifacts resulting either from involuntary patient motion or MRI detection techniques. One problem, that is not yet solved, is related to the dependency of the results in regard to the threshold applied within each analysis. Recently, an alternative fMRI time series analysis method has been proposed to decrease the dependency of the results with the applied threshold. This method is based on the calculation of the generalized mutual information (GMI) between the blood oxygenation level dependence time series and the paradigm slope. Since the analysis relies heavily on the Tsallis q parameter, a broad spectrum of results can be produced. So simulated data analysis was performed in order to construct the receiver operating characteristic curves. The parameters that evaluate the quality of the curves were determined thus allowing for an optimization of the q values. Results for simulated and real fMRI data were compared for both GMI and classical deterministic analysis.