Evaluation of Similarity Measure for Feature based 3D-2D Registration in Vertebral Pose Estimation

The vertebral pose estimation helps to assist the clinician during the surgery to avoid the wrong site surgery. The vertebral poses are estimated by establishing the spatial correspondence relationship between the pre-operative MR image and intra-operative X-ray, and then assessed by a similarity measure. Many similarity measure techniques such as mutual Information, normalized cross-correlation, etc., were used in 3D to 2D registration techniques. So the proposed framework employs three different similarity measures: Binary Image Matching, Dice Coefficients, and Normalized Cross-correlation technique to compare the images based on pixel positions to evaluate the registration performance. In the absence of Intra-operative X-ray images, the experiment is conducted on the simulated test images. Registration accuracy is estimated based on the mean Target Registration Error. The results demonstrated that the three similarity measures work well for feature-based 3D to 2D registration in that the Dice coefficient gives the best result.