Sensitivity of Initial Pose Variations to Feature based 3D-2D Registration in Vertebral Pose Estimation

The vertebral pose is accurately estimated by registering the pre-operative MRI with the intra-operative x-ray during the surgery. The registration task becomes considerably more challenging for multimodal images due to differences in image pixel intensities. The registration algorithm fails when there is a large initial displacement between the images. This paper proposes a feature-based 3D to 2D registration to accurately align pre-operative MRI with intra-operative X-ray with large displacement. The experiment was conducted on the simulated images by varying the initial displacements. The results demonstrate that the proposed method can increase the starting displacement upto a range of ±20mm and ±10° displacements in all translation and rotation parameters, respectively, by providing a high accuracy.