HEARTFUSION: Imaging-driven Patient-specific Cardiac Simulation
Due to aging of our western societies, cardiovascular diseases pose one of the most significant health problems - may it be in terms of the number of people affected, the alarmingly poor prognosis, or the high associated healthcare costs. In 2011, in the USA alone, 813.000 deaths resulted from cardiovascular disease. Clearly, the understanding, diagnosis, and treatment of heart failure are of major importance for our societies and have a strong impact on economics and social life. In this context, mathematical modeling and numerical simulation have become important and indispensable tools also for medical applications during the last decades. In fact, the ever-growing capacities of modern supercomputers and advancement in numerical methods have allowed for the realization of more and more accurate “in-silico” studies of the human heart. Detailed cardiac models are currently employed in modern simulation tools – mostly based on discretization methods such as finite differences or finite elements – to study, e.g. the electrophysiological activation of the human heart, the electro-mechanical deformation, or the fluid flow in the heart and the cardiovascular system. By design, besides medical research, these simulation tools can be used for early diagnosis and patent specific predictive therapy planning. Unfortunately, the time and effort needed for setting up an individualized patient specific simulation based on clinically available data is still prohibitive for using “in silico” analysis as an everyday tool in clinical practice. As a matter of fact, the large set-up times are mostly due to the construction of the geometric models, i.e. the meshes, which are the basis for almost all discretization schemes used in practice. In a standard workflow, the creation of the (surface) meshes involves a semi-automatic segmentation processes, followed by surface creation and surface editing. This process is not only time consuming, but also potentially error prone and definitely tedious. Often, the eventual outcome, i.e. the mesh, only provides an approximation to the segmented geometry, as the complex geometry of the heart is usually approximated by a coarser mesh in order to reduce the number of unknowns. In case of time dependent geometries such as a beating heart, a “reference point in time” must first be defined. For this, one spatial geometry then is reconstructed. From the point of view of numerical simulation and later validation, however, it would be much more desirable to have access to the complete geometry of the beating heart in space and time. It is the goal of the present project HEARTFUSION to address these difficulties by providing an automated approach for personalized simulations in cardiology, which is integrating numerical simulation and imaging data as tightly as possible. In order to achieve this target, we will first construct a space-time (3 + 1D) atlas for the moving heart. In fact, the atlas will be stored by means of discrete representations of diffeomorphisms, which map a carefully constructed and already meshed “mean-value-heart” to the respective patient’s heart geometry. As the information stored in this diffeomorphism already contains all the needed geometrical information, in a next step we will make the information in this space-time atlas directly available to our finite element discretization by using super-parametric elements. In this way, we do not only always compute on the “exact” geometry but we also simplify and shorten the standard workflow for patient specific simulations significantly, as now no individual meshes have to be created. Finally, we plan to make full use of the available space-time geometric knowledge by running simulations for cardiac electrophysiology on the geometries provided through the space-time atlas. The main ingredients of HEARTFUSION will be: • the development of novel space-time capable techniques for atlas based segmentation. • available image data (MRI/CT) (feature point detection and local refinement) in 3 + 1D • integration of the complete geometry knowledge from the geometry reconstruction into the finite element simulations by means of volume parameterizations • adaption of the discretization schemes used in space and time • electrophysiology studies on the space/time deformed configuration as “proof of concept” • the development and implementation of the corresponding software tools.