Heart diseases are major causes of death worldwide. Due to the complex structure of the heart and its interplay with blood circulation, diagnosis and therapy planning for heart diseases are complex tasks. Advances in imaging technologies, such as cardiac computed tomography (CT), cardiac magnetic resonance imaging (MRI), and ultrasound (US), have introduced a range of complementary techniques for analyzing cardiovascular tissue properties, functionality, and hemodynamics. However, the clinical interpretation of the resulting imaging data remains challenging due to the complex three-dimensional anatomy and motion of the cardiovascular system, which are influenced by myocardial performance and the interactions between the ventricles, atria, and vascular systems. To achieve a comprehensive understanding of the heart's condition, complementary functional information, including motion, perfusion, and blood flow, must be integrated. The goal of this thesis is to develop, implement, and evaluate concepts to support collaborative cardiac imaging research, taking into account the special requirements arising from the motion properties caused by the contraction of the heart and breathing. The first part of the thesis describes the design, implementation, and evaluation of the platform for cardiac image processing (CAIPI). This platform supports the multimodal 4D image import, alignment, quantitative comparison of results from different cardiac image acquisitions, and plugin mechanism and comparison of post-processing algorithms for cardiac images. A fully automatic deep-learning-based U-Net segmentation of the right ventricle in cardiac cine imaging is developed to demonstrate the framework's capabilities. The second part introduces a collaborative software environment for developing and evaluating machine learning solutions for image segmentation or disease classification tasks. By integrating visual analytics, machine learning applications can be developed iteratively following an expert-in-the-loop paradigm, shown in two use cases. In the third part of the thesis, we develop a framework for the radiomics analysis in velocity-encoded 4D phase-contrast MRI, which was used to differentiate between healthy volunteers and patients with aortic valve stenosis based on the software environment from the previous parts. In this thesis, we demonstrate through specific use cases and collaboration with external partners that these concepts are actively applied in image-based cardiovascular research.
Markus Hüllebrand (Thu,) studied this question.