In many cases X-ray images are the only basis for surgery planning. Nevertheless it is desirable to draw conclusions about the 3D-anatomy of the patient from such data. This work presents a method to reconstruct 3D shapes from few digital X-ray images on the basis of 3D-statistical shape models. At the core of this method lies an algorithm which optimizes a similarity measure assessing the difference between projections of the shape model and the X-ray images. Based on theoretical and experimental observations we propose to measure the distance between the silhouettes of the object in the projections. The method is tested on 23 synthetically generated X-rays from CT data sets of the geometrically as well as topologically complex shape of the pelvic bone.
This work won the Karl-Heinz Höhne Preis from the GI Special Interest Group MedVis in 2008.