Evaluation of a Statistical Shape Model Based Approach for Recovering the 3D LV Shape from Projective X-Ray Images

Recovering the 3D shape of the left ventricle from contrast-enhanced bi-planar x-ray image sequences is a challenging task. The inherently sparse and noisy data available for reconstruction and the ill-posed nature of the problem necessitate the incorporation of prior knowledge about the ventricular anatomy. Our novel approach reconstructs the endocardial surface from two projections using a statistical shape model that is learned from high-resolution multi-slice CT data. A non-rigid 2D/3D registration method fits the deformable model to the recorded x-ray images by calculating simulated projections of the model and minimizing the difference between simulated and real projections. The presented algorithm is evaluated based on simulated and real patient data using volumetric and geometric similarity measures. For the first time, bi-planar and CT images of the same patient are used to compare the recovered LV shape with the actual shape.