We have developed several methods for automated analysis of echocardiographic images. This thesis describes these methods and their evaluation and use. It is shown that semiautomatic detection... Show moreWe have developed several methods for automated analysis of echocardiographic images. This thesis describes these methods and their evaluation and use. It is shown that semiautomatic detection based on Dynamic Programming and Pattern Matching provides a useful and reliable way of analyzing 2D echocardiographic sequences of different cross sections. Main conclusion is that the new detection tools based on statistical models (Active Appearance Models) provide superior possibilities for automated analysis of echocardiographic images, since they are capable of realistically modeling both the typical problems and artifacts of cardiac ultrasound and the variability between patients. Also, these tools can be extended towards multi-view and multi-stage applications (e.g. stress echo), higher dimensions (3D echo), and simultaneous detection of multiple structures (LV, RV, atria, epicardium, valves). They also offer possibilities for computer-aided diagnosis, such as wall motion abnormality classification (stress echo and Cardiac Resynchronization Therapy). Further development and integration with other border detection and tracking approaches is certainly feasible and will offer a range of new research opportunities. The difficulties have not yet been completely overcome, but we are confident that fully automated, reliable analysis of echocardiographic images will eventually become a definite reality. Show less
