Registration algorithm for statistical bone shape reconstruction from radiographs - an accuracy study.

Medical diagnoses relying on 3D data has become a procedure in clinical practice, increasing the valuable diagnostic information. However, high costs are associated with common 3D imaging modalities and therefore posing the demand for alternative approaches. Correspondingly, 2D-3D registration in terms of deforming a surface model of the anatomy of interest under the constraint of statistical plausibility has become an active field of research. In this context a quantitative analysis of the accuracy of the resulting 3D shape related to a prior established gold standard is necessary, especially since registration is based on sparse information about the patient's true anatomy. The present study discusses validation methods and results. Furthermore approaches for further improvement are provided.