Optimization techniques for semi-automated 3D rigid registration in multimodal image-guided deep brain stimulation

Abstract

Multimodal image registration is vital in DeepBrain Stimulation (DBS) surgery. DBS treats movement dis-orders by implanting a neurostimulator device in the brain todeliver electrical impulses. Image registration between com-puted tomography (CT) and cone beam computed tomography(CBCT) involves fusing images with a specific field of view(FOV) to visualize individual electrode contacts. This containsimportant information about the location of segmented con-tacts that can reduce the time required for electrode program-ming. We performed a semi-automated multimodal image reg-istration with different FOV between CT and CBCT imagesdue to the tiny structures of segmented electrode contacts thatnecessitate high accuracy in the registration. In this work, wepresent an optimization workflow for multi-modal image reg-istration using a combination of different similarity metrics,interpolators, and optimizers. Optimization-based rigid imageregistration (RIR) is a common method for registering images.The selection of appropriate interpolators and similarity met-rics is crucial for the success of this optimization-based imageregistration process. We rely on quantitative measures to com-pare their performance. Registration was performed on CT andCBCT images for DBS datasets with an image registration al-gorithm written in Python using the Insight Segmentation andRegistration Toolkit (ITK). Several combinations of similaritymetrics and interpolators were used, including mean squaredifference (MSD), mutual information (MI), correlation andnearest neighbors (NN), linear (LI), and B-Spline (SPI), re-spectively. The combination of a correlation as similarity met-ric, B-Spline interpolation, and GD optimizer performs thebest in optimizing the 3D RIR algorithm, enhancing the visu-alization of segmented electrode contacts. Patients undergoingDBS therapy may ultimately benefit from this.