A Statistical Approach to Automatic Heart Segmentation and Modelling from Multiple Modalities

In this work, we consider a statistical approach to the fully automatic segmentation of heart walls from magnetic resonance imaging (MRI) data and the reconstruction of three-dimensional models from other modalities. The method is based on utilizing a Markov random fields (MRF) model that provides powerful opportunities for noise suppressing and, at the same time, for an accurate contour preserving segmentation. Implementing local statistics on pixel neighbourhoods makes possible the detecting weak borders, that are even not detected by observation. The process is followed by the active appearance modelling to reduce the oversegmentation. Further, the statistical shape database can be used for 3D reconstruction of complete heart models from data derived by using other image acquisition techniques, like cardiac ultrasound. This approach was evaluated on a set of degraded phantom data.