Combined imaging modalities, like Computer Tomography (CT) and two-dimensional ultrasound, are used to obtain sufficient visual information for minimally invasive procedures, where visibility of organs is typically obstructed by the skin of the patient. The supporting quality of a combined imaging modality can be improved by application of Three-dimensional TransEsophaegal Echocardiography (3D TEE). However, the relatively low signal-to-noise ratio, occurrence of artifacts and the irregular appearance of catheters make the interpretation of the ultrasound volumes a difficult task. Automatic localization of the catheter can help to select the optimal scan plane or region of interest for visualization. In this thesis the feasibility of detection and orientation estimation of a cardiac ablation catheter tip in 3D TEE volumes is studied. An algorithm is developed for automatic localization of the tip of an ablation catheter. First, a tuned vesselness filter from literature is used to enhance the catheter appearance inside the ultrasound volumes. Subsequently, RANdom SAmple Consensus (RANSAC) is applied to detect and fit the tip axis and define its position and orientation in 3D. In experiments we have validated the algorithm in water, a rubber heart phantom and an ex-vivo pig heart. In the heart phantom we achieve an accuracy of about 1.8 mm and 0.08 radians of the position and orientation estimation, respectively (in about 7 seconds). Finally, we explore the application of adding a passive marker to the catheter tip for roll estimation. We find that application of passive markers for roll estimation is attractive, but requires sufficiently high-resolution imaging.
|Begeleider||Peter H.N. de With (Afstudeerdocent 1), Sveta Zinger (Afstudeerdocent 2), Fei Zuo (Externe coach) & Alexander F. Kolen (Externe coach)|