A statistical shape model of the heart and its application to model-based segmentation

In the present paper we describe the automatic construction of a statistical shape model of the whole heart built from a training set of 100 Multi-Slice Computed Tomography (MSCT) studies of pathologic and asymptomatic patients, including 15 (temporal) cardiac phases each. With these data sets we were able to build a compact and representative shape model of both inter-subject and temporal variability. A practical limitation in building statistical shape models, and in particular point distribution models (PDM), is the manual delineation of the training set. A key advantage of the proposed method is to overcome this limitation by not requiring them. Another one is the use of MSCT images, which thanks to their excellent anatomical depiction, have allowed for a realistic heart representation, including the four chambers and connected vasculature. The generalization ability of the shape model permits its deformation to unseen anatomies with an acceptable accuracy. Moreover, its compactness allows for having a reduced set of parameters to describe the modeled population. By varying these parameters, the statistical model can generate a set of valid examples. This is especially useful for the generation of synthetic populations of cardiac shapes, that may correspond e.g. to healthy or diseased cases. Finally, an illustrative example of the use of the constructed shape model for cardiac segmentation is provided.