Pre-alignment for Co-registration in Native Space

Abstract

For nonlesional patients, the correct localization of the epileptogenic foci in native space remains a great challenge. Non-invasive functional PET images that provide information about cerebral activities may reveal the origin of seizure activity, but without precise anatomical detail. Co-registration of the functional images with MR images on the basis of maximization of mutual information (MMI) has shown to be very promising in improving presurgical evaluation. Nevertheless, a mutual information (MI) function is non-convex and the convergence of an algorithm to its optimum is guaranteed only if the initial estimate lies in its convex vicinity. We present in this paper a generally applicable method that pre-aligns the DICOM images such that their relative position becomes close to an optimum. The key to our solution is a robust user-guided interactive procedure to extract valid voxels, for both the centroid estimation and the registration. Aiming at comparative analysis, we introduce a numerical condition to quantify registration errors. The results are acceptable when we consider the intrinsic problems of the MMI-based registration algorithm we implemented.