From medical images to individualized cardiac mechanics: A physiome approach.

摘要

Cardiac mechanics is a branch of science that deals with forces, kinematics, and material properties of the heart, which is valuable for clinical applications and physiological studies. Although anatomical and biomechanical experiments are necessary to provide the fundamental knowledge of cardiac mechanics, the invasive nature of the procedures limits their further applicability. In consequence, noninvasive alternatives are required, and cardiac images provide an excellent source of subject-specific and in vivo information.;Noninvasive and individualized cardiac mechanical studies can be achieved through coupling general physiological models derived from invasive experiments with subject-specific information extracted from medical images. Nevertheless, as data extracted from images are gross, sparse, or noisy, and do not directly provide the information of interest in general, the couplings between models and measurements are complicated inverse problems with numerous issues need to be carefully considered.;The goal of this research is to develop a noninvasive framework for studying individualized cardiac mechanics through systematic coupling between cardiac physiological models and medical images according to their respective merits. More specifically, nonlinear state-space filtering frameworks for recovering individualized cardiac deformation and local material parameters of realistic nonlinear constitutive laws have been proposed.;To ensure the physiological meaningfulness, clinical relevance, and computational feasibility of the frameworks, five key issues have to be properly addressed, including the cardiac physiological model, the heart representation in the computational environment, the information extraction from cardiac images, the coupling between models and image information, and also the computational complexity. For the cardiac physiological model, a cardiac physiome model tailored for cardiac image analysis has been proposed to provide a macroscopic physiological foundation for the study. For the heart representation, a meshfree method has been adopted to facilitate implementations and spatial accuracy refinements. For the information extraction from cardiac images, a registration method based on free-form deformation has been adopted for robust motion tracking. For the coupling between models and images, state-space filtering has been applied to systematically couple the models with the measurements. For the computational complexity, a mode superposition approach has been adopted to project the system into an equivalent mathematical space with much fewer dimensions for computationally feasible filtering. Experiments were performed on both synthetic and clinical data to verify the proposed frameworks.