Sub-Voxel Unwarping of Spin Echo EPI to Conventional Anatomical MRI

Purpose This work examines an extension of an entropy based non-rigid registration algorithm to recover very localised sub-voxel distortion of EPI data relative to high-resolution anatomical MRI. Such a registration estimate can then be used to provide increased accuracy of the anatomical localisation of fMRI activations. Introduction Accurate alignment of fMRI data with high-resolution anatomical MRI data can significantly aid the interpretation of fMRI activations. Magnetic field inhomogeneities can introduce significant localised distortion in EPI data acquired for fMRI studies [1]. In earlier work [3] we have developed the approach of introducing an imaging distortion model into the non-rigid registration of spin echo EPI with anatomical MRI data. In this work we have examined an extension of the approach using image sub sampling to recover the most localised distortions, which can occur in the temporal lobes. Methods Following the approach of [3], we make use of the spin echo signal conservation property as a constraint in the nonrigid registration and intensity correction of the EPI data to the anatomic MRI. The Jacobian of the geometric transformation thus modulates the intensity of the EPI registration estimate [2]. We employ normalised mutual information [4] as an alignment criterion between T1 weighted anatomical MRI values, A, and the logarithm of corresponding corrected spin echo EPI values ET,