Bone tracking from X-Ray sequences

Bone tracking from X-Ray sequences is important for biological and evolutionary applications like motion analysis of species, and also in human health, for example in surgical guidance, study of kinematics, or post-operative implant assessment. Since X-Ray data is 2-dimensional, it does not provide enough information about the 3D position of the bones. C-Arm technology (Fig. 1) reduces this problem by capturing two simultaneous views of the bone. Using this stereo vision and a volumetric model of the bone, typically a CT scan, the search is performed by comparison of a synthetically computed view (Digitally Reconstructed Radiography or DRR) and the X-Ray image. The objective is to solve for the transformation of the bone that produced each X-Ray image. The high density of the CT scan data makes DRR synthesis a computationally expensive task. Also, high frame rate in the sequences is desired for tracking accuracy, but it results in noisy and low contrast X-Ray images [1], which makes the registration difficult. Furthermore, comparing multi-modal data like in the case of X-Ray and CT is challenging because they don’t follow linear transformations [2]. In this project we explore three techniques, that address these issues.