Contributions to assessment of cardiac failures by echocardiography image processing

This thesis focuses on the analysis and processing of 3D echocardiographic imaging, being its aim to provide technical support to accurately assess different myocardial failures from ultrasound data of the heart. Results provided are sought to be robust, independent of the intra- and inter-observer variability common in radiological interpretation.

Specifically, the thesis focuses on two main tasks: (1) the characterization of the dynamic properties of the left ventricle and (2) the detection of different structures therein. For this purpose, methods for the estimation of motion and strain are proposed, as well as a method for identifying the different cardiac structures, such as the mitral valve, the long axis of the left ventricle and the aortic valve. For the sake of comprehensiveness, we introduce a thorough review of image registration techniques, which surveys the theoretical foundation of the models and classifies them according to this basis. Alongside, a review of the most relevant speckle tracking techniques and an extensive analysis of the influence of the different speckle tracking strategies in motion and strain estimation is carried out.
The methodology used for detecting the salient cardiac structures is based on  probabilistic and structural tissue characterization, the Hough transformation for circles and multidimensional dynamic programming. This approach provides a robust and accurate detection to initialize multicavity segmentation techniques without any user interaction. On the other hand, the main contribution concerning image registration is twofold: first, the inclusion of accurate models for speckle, using a free-form deformation and a diffeomorphic diffusion model; second, the influence of the speckle tracking strategies --such as the use of different speckle, transformation and optimization models-- in the motion and strain accuracy is extensively studied. The proposed study also takes into account the improvement of the accuracy of the method when myocardial structural information is included into the speckle tracking techniques.

Finally, this thesis also has relevant practical contributions to be used in clinical practice. Throughout the thesis, we show and emphasize the importance, the potential and the relevance of the strain to quantify cardiac function and assess different pathologies, such as myocardial ischemia, functional or ischemic mitral insufficiency, and dyssynchrony. Therefore, an accurate estimation of such measure becomes essential. We propose different schemes for (1) unsupervised detection of cardiac structures; (2) valvular tissue classification and (3) estimation of myocardial deformation in ultrasound images.