Cross-Validation of Deformable Registration With Field Maps in Functional Magnetic Resonance Brain Imaging

Abstract

The localization of brain functional activity with respect to brain anatomy requires registration between a functional image and a reference high-resolution anatomical image. The fast functional magnetic resonance brain images acquired via echo planar imaging (EPI) in functional magnetic resonance imaging (fMRI) suffer from local geometric distortions. After performing standard rigid or affine registration, local nonlinear distortions of up to 10 mm still appear in prefrontal and orbitofrontal cortex, as well as in temporal and cerebellar regions. To reduce such distortions, an enhanced version of a previously developed deformable registration technique is introduced in this paper. The enhanced deformable registration technique involves a multistage mutual information-based free-form deformable framework guided by a field map template. This template is utilized as prior knowledge for performing an initial warp as well as a guiding mechanism to constrain the initial stage of the deformable registration. The enhanced technique is cross-validated with the high-resolution field map distortion correction technique using real datasets. The results from consistency tests using a normalized mutual information similarity measure as well as the statistics of quantitative cross-validation show the degree of enhancement achieved and the impact of the developed deformable registration in applications requiring accurate local registration.