SU‐GG‐I‐106: Evaluation of Free‐Form B‐Spline Deformable Registrations for Noisy CBCT/SimCT Images

Abstract

Purpose: Accurately registering noisy cone‐beam CT (CBCT) images to simulation CT (SimCT) is a challenging task. The goal of this study is to evaluate B‐spline based free‐form deformable registrations with noisy CBCT images. Method and Materials: CBCT images were retrospectively selected with large rectum gases and/or significant motion artifacts around bones. Then, they were registered to corresponding SimCT images using Elastix, a software package implemented in conjunction with ITK (Insight Segmentation and Registration Toolkit). Each registration consisted of three sequential stages; translation, affine, and B‐spline. Each stage had three resolutions to avoid local minima and to reduce the overall registration time. The B‐spline grid spacing in the finest resolution was (8, 8, 2) voxel in the (x, y, z) directions respectively. Mutual information (MI) was chosen as similarity metric with random sampling per iteration. The sampling size was 10% of the image size. Registrations were evaluated using an interactive blending software. Results: External skin boundaries were found to match well, but internal organ registration included notable errors, especially near areas containing large amount of rectum gas. Errors were also observed around the top and bottom slices of the input CBCT images, where the bones were shown to “stretch” in the superior/inferior direction. Conclusion: Based on the preliminary results, the metric MI with free‐form deformation alone is inadequate for noisy CBCT/SimCT registrations. Our future study includes using rigidity constraints on bones and customary constraints on deformable organs.