3D cardiac anatomy reconstruction using high resolution CT data

Abstract

Recent advances in CT technology have allowed the development of systems with multiple rows of detectors and rapid rotation. These new imaging systems have permitted the acquisition of high resolution, spatially registered, and cardiac gated 3D heart data. In this paper, we present a framework that makes use of these data to reconstruct the 3D cardiac anatomy with resolutions that were not previously possible. We use an improved 3D hybrid segmentation framework which integrates Gibbs prior models, deformable models, and the marching cubes method to achieve a sub-pixel accuracy of the reconstruction of cardiac objects. To improve the convergence at concavities on the object surface, we introduce a new type of external force, which we call the scalar gradient. The scalar gradient is derived from a gray level edge map using local configuration information and can help the deformable models converge into deep concavities on object's surface. The 3D segmentation and reconstruction have been conducted on 8 high quality CT data sets. Important features, such as the structure of papillary muscles, have been well captured, which may lead to a new understanding of the cardiac anatomy and function. All experimental results have been evaluated by clinical experts and the validation shows the method has a very strong performance. © Springer-Verlag Berlin Heidelberg 2004.