Building a continuous bone surface representationFrom the CT gray-scale volume images we get 3-D points corresponding to the bone cortex (via thresholding, edge detection, contour grouping, and massive user interaction). 'Cortex' is a fancy word for 'shell' (bones have some soft tissue on the inside).From a 3-D point cloud we build a smooth, locally parameterized, manifold surface. In order to speed up the reconstruction process we are reusing the manifold representation of a specific bone from one patient to another patient. White - cloud of 3D points (meshified for visualization purposes). Pink - canonical model of the bone (manifold); the canonical model deforms to fit the bone of a new patient. Top row: a) The position of the bones in the CT scans differs from patient to patient. b) We have an automated procedure for aligning the bones. c) The canonical model deforms to fit the new bone (shifted to the side for viewing only). Bottom-row: CT-scans of the forearm are not complete (i.e., some forearms are scanned down to the elbow, some just a little below the wrist). The alignment procedure still works. So does deforming. 
|
