A methodological approach to the study of periodically deforming anatomical structures

We present a methodology, based on neural paradigms, suitable to analyze periodically deforming anatomical structures and recognize their state. Anatomical structures, considered as 'multimedia objects', are defined as organized sets of images and signals, acquired from multiple sources. These sets are combined and processed using dedicated Artificial Neural Networks to obtain a 3D reconstruction of the object, at different times of its dynamic evolution (deformation cycle). In order to reduce acquisition errors, we consider also an inter-cycle registration of the volumes, resulting in the most likely 3D reconstruction. Morphological and functional characteristics are then associated to each element of the reconstructed volume and processed to discriminate different object states. The developed methodology has been applied to analyze cardiac dynamics and, in particular, to identify physio-pathological states of the left ventricle.