Ultraschallbasierte Navigation für die minimalinvasive onkologische Nieren- und Leberchirurgie

In minimally invasive oncologic renal and liver surgery with many advantages for the patient, the surgeon is often confronted with orientation problems. Main causes are the indirect view of the patient's anatomy, the limited field of view and the intraoperative deformation of organs. Navigation systems, which are often based on intraoperative ultrasound, can be a solution. With real-time imaging, the deformation of the organ can be determined. Since many tumors are not visible in the ultrasound picture, robust and automatic deformable registration with the preoperative CT is required. Furthermore, a permanent visualization during the manipulation of the organ is necessary. For the kidney, the suitability of ultrasound-elastography-imaging for image-based registration was evaluated using Mutual Information. However, this had only moderate success, due to poor image quality and the small field of view of the image data. The branches and endpoints of blood vessels of the liver can be used as natural landmarks for the registration process. In a first step, vessel segmentation algorithms have been developed for the two most common types of ultrasound imaging B-mode and power Doppler. The proposed combination of the two modalities increased the amount of vessel branches on average by 35 %. For the rigid registration of vessels from ultrasound and CT on average 9 bijective point correspondences are defined using an existing graph matching method [OLD11b]. The average registration accuracy is 3.45 mm. The amount of point correspondences is not sufficient for a deformable registration. The developed method for a landmark refinement adds more points between matched landmarks along the vessel center lines and searches for other corresponding vessel segments increasing the average number of point correspondences to 70. This allows the determination of the organ deformation based on the differences in the vascular orientation. Using thin-plate splines, a deformation field can be computed for the entire organ, based on the point correspondences. In this way, the registration accuracy is increased on average by 44 %. The most important prerequisite for the success of the deformable registration is an extensive vessel segmentation from the ultrasound image data. As part of this work, the concept of Regmentation was extended to the segmentation of vessels and vessel-based registration for the first time. Through the combination of both methods, the extracted vessel length was increased by a mean of 32 %, resulting in an increase of the average number of corresponding landmarks to 98. By this means, the deformation of the organ, and thus the change in position of the tumor, can be specified more accurately and with greater confidence. With the knowledge of the tumor location in the organ and by use of a marker wire the change in position of the tumor during surgical manipulation can be monitored using an electromagnetic tracking system. By this tumor tracking a permanent visualization using video overlay in laparoscopic video images becomes possible. The main contributions of this work for vessel-based registration are the vessel segmentation from ultrasound image data, the landmark refinement to obtain a high number of bijective point correspondences and the introduction of Regmentation to improve vessel segmentation and deformable registration. The tumor tracking for navigation enables permanent visualization of the tumor during the entire procedure