3D + t Statistical Human Heart Phantom for X-Ray Projection Imaging

The long acquisition times of about 5s in rotational angiography using C-arm CT systems requires the incorporation of the heart’s dynamics into the reconstruction algorithms. Meaningful evaluation and comparison of such algorithms requires projection data from computerized 4D phantoms. Changing clinical parameters of the model or incorporating pathologies, however, may decrease clinical relevance. In order to tackle this problem, we develop a 4D open-source statistical shape model of the whole heart.

We compare two registration-based segmentation based on Thirion’s Demons and mutual information driven B-spline transforms, respectively, for automatic training set generation from four-chamber enhanced ten phase CT angiography. The model, built from nine male and eleven female patient’s data sets, exhibits a generalization ability of 5:00  0:93mm and specificity of 7:30  0:97mm. We investigate the use of kernel principal component analysis in classification of pathologies with manifestations in anatomical alteration.

The model is implemented in CONRAD, an open-source simulation and reconstruction framework. The algorithms and high resolution projection data are available online. To the best of our knowledge, this is the first open-source statistical shape model of its kind.