Bilinear point distribution models for heart motion analysis

This paper presents a cardiac motion modeling method that separates motion from anatomical patient specific cardiac features by using bilinear models on segmented heart volumes. This factoring in a quasi-independent anatomy-motion transform significantly reduces the dimensionality of the data to analyze: instead of having to check all the points on the volumetric mesh, we analyze a reduced number of parameters corresponding to the shape variation induced by motion. Bilinear models are first learnt in a large population of subjects, and then applied individually to each subject to derive a patient specific model. This provides information about individual specific motion parameters, facilitating the detection of individuals that are far from the mean cardiac motion pattern and motion interpolation.