Content area
Abstract
3D personalized biplane reconstructions of scoliotic rib cages are frequently performed at the Sainte-Justine Hospital in Montreal, Canada. The use of a pair of digital radiographs makes it possible to avoid high levels of radiation and other problems related to classical 3D imaging modalities such as CT-Scan for instance. However, the reconstruction algorithm (Dansereau and Stokes, 1988; Marcil, 1996) requires a nonstandard X-ray PA-20° and the extensive, very time-consuming (on average, two hours per rib cage) manual positioning of 440 specific landmarks in the images.
In this project, the 3D/2D registration technique suggested to reconstruct scoliotic rib cages relies on the work done by Benameur and al. (2001, 2003) on scoliotic vertebrae and uses only conventional x-ray images PA-0° and LAT. The reconstruction of the complete rib cage is carried out in a hierarchical way: each reconstructed rib determines the initial pose parameters (size and orientation) of the following one. The 3D reconstruction of each rib is based, through the analysis of the data base of the Sainte-Justine Hospital, on the development of a set of 3D statistically deformable templates (Cootes and al., 1992) of rib centerlines and meshed models of their surface. Rib edges are automatically extracted from the postero-anterior radiograph using the Canny algorithm (1986) and post-processing steps involving thresholding and linking of the edges. In the lateral view, reduced visibility prevents from automatic treatment and thus, a set of 60 points are manually digitized. The 3D/2D registration problem is stated as the optimization of a cost function measuring the quality of the fit between the data extracted from the radiographs and the projections of the deformable models in both images. (Abstract shortened by UMI.)
