engleski

Model-based segmentation and texture analysis of multi-modal cardiac images

In view of prevalence and fatality of cardiac diseases in the modern world, cardiac and cardiovascular imaging modalities have become indispensable tools for clinical patient examination. They allow for visualising internal cardiac structures that are otherwise inaccessible, recognising their substructures, simplifying their visual representations and quantifying their physical and physiological characteristics. However, limitations of imaging systems and built-in algorithms for analysis of imaged data relate to an incomplete understanding of all causal links in the cardiovascular system. With the ongoing advancements in imaging technologies, experimental setups and algorithm design, there is an opportunity for clarifying the underlying cardiovascular processes more precisely. This thesis presents computational methods for analysis of different types of images that capture contraction-related scale-specific phenomena in the cardiovascular system. The analysis begins with quantification of subcellular deformation from high-resolution confocal microscopy images of isolated cardiac muscle cells during contractions. The developed algorithm combines the cross-correlation based approach with a set of mathematically formulated restrictions that regularise intracellular motion. At mid-scale level, the arrangement of myocytes within the cardiac tissue is analysed from phase-contrast micro-CT cardiac volumes. The applied method consists of estimating fibre vectors using eigen-decomposition of local gradient structure tensors, and quantifying fibre orientations in locally defined coordinate systems. At the highest level of the approach, coordinated contractions of fibres cause pumping of the blood out of the heart through the aortic valve to aorta and the rest of the body. Delineation of blood flow velocities from aortic Doppler blood flow images is achieved by coupling mathematical modeling of circulation with a procedure for atlas-based image segmentation. The described flow of research corresponds to a gradual change in imaging resolution and represents a multilevel and multimodal approach to image-based analysis of the cardiovascular system. Throughout all levels of analysis, artificial cardiac images and textures were synthesised with respect to modality-specific properties and analysed in parallel with the acquired images. This guarantees that the data of interest is balanced and representative in terms of imaged content diversity, which leads to an increase in performances of algorithms designed for their analysis.

cardiac image analysis ; mechanical deformation ; structure tensors ; circulation modeling

nije evidentirano