Spatial alignment of functional regions in fMRI

Abstract

An essential step for discovering a common structure in brain activation regions from multi-subject fMRI data is the ability to find spatial correspondences across subjects. This has proven to be a challenging problem due to the lack of a ground truth and variability in anatomical brain structure, functional activation, and spatial locations of functional regions. Standard methods rely on the correspondences given by anatomical registration to a common space, but fail to account for spatial variability of the functional regions relative to anatomy. We develop a clustering method that relies on the alignment of both the anatomical structure and the functional landmarks. The method is shown to improve over standard group analysis techniques that rely on anatomical alignment only. The validation of our method confirms that peaks of activation exhibit consistent spatial structure. Furthermore, our work creates a framework for future testing of different metrics for similarity of brain activation regions across subjects.