Characterizing fMRI activations within regions of interest (ROIs) using 3D moment invariants

A novel method is proposed for characterizing spatial distribution changes in functional magnetic resonance imaging (fMRI) activation statistics under different experimental conditions. The proposed technique, based on three dimensional (3D) invariant moment descriptors, was applied to fMRI data recorded from eight healthy subjects performing internally or externally-cued finger tapping sequences. Voxel-based activation statistics were characterized in several regions of interest (ROIs), including the supplementary motor area (SMA), cerebellum, primary motor cortex, prefrontal cortex, and caudate. Examining the activation patterns of these neural regions using 3D moment invariants revealed that the patterns of activation regions were significantly different during externally and internally cued tasks when computed across subjects. In contrast, traditional methods that are based on amplitude of the activation statistics demonstrated reduced discriminability. The results suggest that the spatial distribution of activation is a more sensitive measure of activation changes, and complements conventional fMRI analyses.