Robust Automated Registration of Wrist Bones Using Tissue Classication and Distance Fields

Abstract We present an intra-subject method for automated registration of wrist bones from computed to-mography (CT) volume images. The method makes possible the automated extraction of joint kinematicinformation from sets of volume images of wrists in different poses. The images may be acquired atrelatively low resolution, reducing the total acquisition time per human subject and general scan storagerequirements. The method is likely to be applicable to bones of other joints.Our registration method works on a sequence of CT images of the same wrist in two steps: in therst step, each CT volume is processed by a tissue classier that generates a localized distance eld.The distance eld approaches zero at bone boundaries. In the second step, the distance elds and ageometric bone model are used to track bone motion through the sequence of CT images. The distancevalue of each vertex in the geometric model is looked up in the corresponding distance eld. The boneposition and orientation are then automatically adjusted to minimize these distances. We infer a motion-governing bone hierarchy from the distance information. The derived hierarchy may have biologicalsignicance. We use the joint hierarchy to speed up the registration process.We validate the method on synthetic,