Page - 46 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

Figure 1. Biplane cine-angiographic x-ray equipment used in the catheter lab to acquire images for quantitative leftventricleanalysis. The gold standard for quantitative left ventricle analysis in the catheter lab is based on the evaluation of end-diastolic (ED) and end-systolic (ES) endocardial contour information gathered from these 2- D projection images. The ED and the ES volume are calculated (by applying e.g. the Area-Length method) and used to determine ejection fraction (EF), i.e. the volume that is squeezed out during contraction. Contour information is further utilized by wall motion analysis methods (like e.g. the Centerline method) to quantify myocardial viability. However, since 3-D information is lost due to projection, volumetric diagnostic parameters, like EF, can only be approximated and wall motion is only evaluable for LV surface areas with the boundary visible in the projection image. Novel approachesaimat reconstructing thespatio-temporalshapeof theLVtoperformanalysis in3-D[10]. 2. RelatedWork Inclassical computed tomography(CT),hundredsofprojectionsareacquiredbya fast rotatingx-ray gantry. Analytical and algebraic reconstruction techniques exploit this dense information to yield voxel values that vary within a continuous range. However, these techniques typically fail if merely two (noisy) projections are available. C-arm CT is a relatively young and hybrid type of imaging modality, where the C-arm is rotated during acquisition to increase the number of projections. Tech- niquesknownfromCTcan thenbeutilized toaddress the reconstructionproblem[8]. In thecatheter lab, however, the application of C-arm CT is challenged by the higher amount of x-ray dose and bolus compared with conventional x-ray angiography (XA), and the slower rotational speed of the C-arm compared with classical CT when imaging the rapidly moving heart. Whether C-arm CT will substituteXAas a routinemethod in future remains to beseen [9]. Unlike classical (continuous) CT, discrete tomography focuses on reconstruction problems where only a small number of projections – as small as two – are available and the object’s intensity levels are limited, i.e. discrete, and known a-priori [3]. Using additional a-priori information is crucial whentrying tosolvesuchunder-determinedandambiguousproblems, since thiscanreduce thespace of possible solutions and improve the ability to deal with noisy projection data. Some of the early approaches published in the field of 3-D LV shape recovery from XA rely on the assumption that ventricular cross-sections follow certain geometric priors (like connectedness, convexity, symmetry, roundness,etc.),however, this isusually toorestrictiveinpractice. IntheworkofPrauseandOnnasch [7], digitized post-mortem human LV casts are used as a-priori information. Other approaches often donot incorporateanatomical a-priori information at all [5], [6]. 46