ABSTRACT
In this work, we propose a novel registration method that models materials of different composition and therefore differing laws of deformation in each of the sub-structures composing the image, and the physical constraints of the deformation that must occur across sub-structures. We accomplish this by automatically detecting (segmenting) sub-structures using their discriminating dynamics or deformation. The deformation in each sub-structure is simultaneously computedwhile the sub-structures are estimated. Further, physical constraints across sub-structures are naturally incorporated into a unified optimization framework for the deformation and the substructure segmentation. Although our framework is valid in a number of registration scenarios (e.g., brain
registration), we demonstrate the idea on the analysis of cardiac MRI, of critical importance [14, 2, 12], which includes the detection of the left ventricle boundary, its deformation, and the deformation in the surrounding area, i.e., the myocardium. For simplicity, we illustrate the formulation on an image comprosed of two sub-structures, although the formulationmay be similarly extended to any number of sub-structures.
