An iterative method to register MR and ultrasound images for MRSI guided prostate implant planning

Abstract

Magnetic Resonance Spectroscopy Imaging (MRSI) has been used to identify the high tumor burdens for patients with prostate cancer. We have conducted a clinical trail that preferentially escalated dose to the regions of high tumor burden identified with MRSI. The patient setup during MRSI and the implant are different and the prostate is deformed during imaging by the endorectal coil and the coupling balloon necessary for the MRSI procedure. As the result, we could only qualitatively identify the region of high tumor burden from the ultrasound image sets and plan more seeds in that region to boost the dose there. The purpose of this work is to develop a deformable image registration method that can fuse the MR and ultrasound images so that the region of high tumor burden identified by the MRSI can be accurately reproduced in the ultrasound image set. An iterative method was used to directly register the deformed MR images to the transrectal ultrasound (TRUS) images for prostate implant planning. This method is a two-level, multi-resolution approach. In the low level, it is a non-linear registration with a quick block-matching method by using the following criteria as the cost function where xi and yi are the i-th voxel in block X of TRUS volume and Block Y of MR volume, respectively. The square error is used to measure the similarity of the two blocks. The rest on the right side is to constrain the displacement vector for smoothing purpose. α is used to balance the two items. In the high level, parameters such as the resolution, block size and filtering function are adjusted. In adjusting the parameters, attentions are also paid to avoid local minima in the low-level registration process. The image set resulted from the last iteration is used as the source image of the next iteration for further refining the match to the target images. The reason for adopting this strategy is that for large non-linear displacements such as in the case between ultrasound and MR images, single step registration could cause unnatural deformations in the prostate. In this preliminarily study, the prostate contours were drawn by a physicist on both image sets. For the MR images, the resolution is 0.63 mm × 0.63 mm × 3.5 mm and there are total 16 slides. For the TRUS images, the resolution is 0.2 mm × 0.2 mm × 5.0 mm and there are total 9 slides. Before registration, the MR image set was resliced, so that both image sets have the same resolution. From the prostate contours, mask images (see figure) were created as the working volumes. The total registration time for 2 typical image sets is less than half hour. The figure shows the two-slide superposition of TRUS (white) and MR (Green) mask images before and after registration. The figure shows the large displacement before registration and these non-linear deformations was reduced to an acceptable degree after registration. An iterative method has been developed to register MR and TRUS images for MRSI guided prostate brachytherapy planning. Initial tests showed the feasibility.