Invariant 3D spharm features for characterizing fMRI activations in ROIs while minimizing effects of intersubject anatomical variability

Spatial patterns of activation statistics within anatomically-defined regions of interest (ROIs) in functional magnetic resonance imaging (fMRI) data were recently shown to be sensitive markers of brain activation changes. Most current methods that analyze fMRI activation statistics largely ignore this. The accuracy and validity of the prevalent approach of spatial normalization of functional data is also being debated. In this paper we present a novel spherical harmonics based rotational, translation and scale invariant feature representation of fMRI data which allows for direct quantification of activation patterns within ROIs without any need for spatial normalization. We also present a novel parallel technique for quantifying anatomical properties of the ROIs where we employ a principal component based approach to reduce the effects of anatomical variability in the ROI on functional pattern analysis. We validate our proposed method and demonstrate its improved sensitivity over conventional methods using real and simulated fMRI data.