Post-implant CT-MR prostate image registration using feature line parallelization and normalized mutual information

Abstract

Post-implant dosimetry for permanent prostate brachytherapy is typically performed using CT images, for which the clear visualization of soft tissue structures is problematic. Registration of CT and MR image volumes can enable improved definition of all structures of interest (soft tissues, bones and seeds) in the joint image set. In this paper we describe a novel two-stage rigid-body registration algorithm that consists of: (1) parallelization of straight lines fit to image features running primarily in the sup-inf (Z) direction, followed by (2) normalized mutual information (NMI) registration. The first stage serves to fix rotation angles about the ant-post (Y) and left-right (X) directions, and the second stage determines the remaining Z-axis rotation angle and X, Y, and Z translation values. The new algorithm was applied to CT and 1.5T MR (T2-weighted and B-FFE sequences) axial image sets for three patients acquired four weeks after prostate brachytherapy using 125I seeds. Image features used for the parallelization stage one were seed trains in CT and needle tracks/seed voids in MR. Simulated datasets were also created to further investigate algorithm performance. Clinical image volumes were successfully registered using the two-stage approach to within an RMS distance of