Page - 102 - in Joint Austrian Computer Vision and Robotics Workshop 2020

ACenterline-GuidedApproachforAortaandStent-Graft Segmentation BertramSabrowsky-Hirsch,StefanThumfart,RichardHofer,WolfgangFenz, ResearchUnitMedical Informatics,RISCSoftwareGmbH,JohannesKeplerUniversity,Austria {bertram.sabrowsky-hirsch,stefan.thumfart,richard.hofer, wolfgang.fenz}@risc-software.at PierreSchmit,FranzFellner CentralRadiology Institute,KeplerUniversityHospital,Austria {pierre.schmit,franz.fellner}@kepleruniklinikum.at Abstract. Monitoring of patients after Endovascu- lar aortic repair (EVAR) is a clinical necessity due to the high re-intervention rate associated with the treatment. The risk assessment could be greatly en- hanced by the inclusion of metrics based on the aor- ticblood-flowandstent-graftchanges. Apreliminary step to this endeavour is, however, the automatic re- constructionof therelevant structures: aorticblood- lumen and the stent-graft wire frame. In this paper we present a centerline-guided approach that lever- ages knowledge about the target structures through a combination of two 3D U-Nets for efficient auto- mated segmentation of both structures. We evaluate our approach on a real-world clinical dataset yield- ingDicesimilaritycoefficientsof0.942and0.841for the blood lumen and stent-graft metal wire, respec- tively. 1. Introduction The abdominal aorta is the largest artery in the human body, with the descending branch supply- ing the lower body with about 4 liters of blood per minute [2]. Abdominal Aortic Aneurysms (AAAs) are critical as a rupture causes massive blood loss that quickly leads to death at a mortality rate of85% to 90% [11], with half of the patients succumbing before they reach a hospital [1]. Overall, AAAs ac- count for 175000 deaths per year globally [7]. In contrast to open surgery, endovascular aortic repair (EVAR) poses a minimally invasive alternative that significantly reduces the intraoperative stress on the patients, who in turn experience shorter periods of (a) (b) Figure 1. An abdominal CT-A scan (a) and a close- up view of imaging artifacts caused by the stent-graft wireframe(b). ThewireframeoftheMedtronicEndurant stent-graft encompasses the blood lumen in the two iliac bifurcations and is itself surroundedby the thrombosis. convalescence. As a result, EVAR is the treatment of choice for60%of patients [3]. These advantages come, however, at the cost of a high re-intervention rate of 20% [18], necessitating post-operative mon- itoring of the patients. We seek to aid monitoring by automatically calculating risk factors from blood flow simulations, which require prior segmentation of the target structures. In this paper, we present a novel method for segmenting the aortic blood lumen andthestentgraftwireframefrompost-operativeab- dominalCT-Ascans. 2.RelatedWork Blood vessel segmentation is an active field in re- search [19] and the variety of approaches reflects the diversity of both the targeted anatomical regions and the available imaging modalities. For clinical monitoring of the abdominal aorta after EVAR, CT- A is the modality of choice [22]. However, unique 102