TY - JOUR T1 - Hepatic arterial tree segmentation: Towards patient-specific dosimetry for liver cancer radioembolization JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 122 LP - 122 VL - 60 IS - supplement 1 AU - Amirtaha Taebi AU - Bahman Roudsari AU - Catherine Vu AU - Simon Cherry AU - Emilie Roncali Y1 - 2019/05/01 UR - http://jnm.snmjournals.org/content/60/supplement_1/122.abstract N2 - 122Objectives: Liver cancer is among the deadliest cancers and its mortality rate is increasing in the United States [1]. Yttrium-90 (Y-90) radioembolization is being increasingly used for the treatment of advanced liver cancer. Accurate pretreatment dosimetry is necessary to determine the Y-90 activity to inject in order to maximize the dose to the tumor while limiting the dose to surrounding healthy parenchyma. Current dosimetry methods are not accurate nor precise, because they do not consider the non-uniform Y-90 microsphere distribution in the hepatic arterial (HA) tree as well as HA anatomy variations among the patients [2]. We are addressing this problem by developing a patient-specific approach combining computational fluid dynamics (CFD) simulation carried out for each patient’s anatomy and Y-90 microsphere decay physics to estimate the amount of dose delivered to different liver tissues more accurately. This study aims at providing a pipeline for extracting HA geometry from cone beam computed tomography (CBCT) images as the first step of our dosimetry approach. Methods: A total of two exams from liver cancer patients were included in this retrospective study. The first subject had multiple hypervascular tumors mainly in the right lobe of the liver. The second patient had a hypermetabolic tumor involving the majority of right hepatic lobe. Breath-hold CBCT was performed using a Siemens Artis Zeego angiography system (6s, 198° coverage with a 0.5° angular sampling). The images were converted into 3D axial images using syngo DynaCT which included a total of 185 1024×1024 images (pixel size of ~0.25 mm) covering the abdominal region. After selecting the volume of interest and performing simple thresholding of the raw images, a fast-marching method was used to extract the closed isosurfaces at a specific gray level of 250-315. The largest connected region was then found to identify the HA tubular surface and remove other unwanted detected regions. The HA surface was smoothed in 30 iterations using Taubin’s algorithm with a passband of 0.01. The Voronoi diagram of the arteries was then used to calculate the centerlines and largest inscribed spheres. The radius of these spheres was used to create a tubular surface around the centerline trajectories. The analysis was done in Matlab and Python library vmtk. Results: Subjects angiograms are shown on the left side of Fig. 1 where the common, proper, right, and left hepatic arteries (CHA, PHA, RHA, and LHA) as well as gastroduodenal artery (GDA) are labeled. The segmented HAs are shown on the right side of Fig 1. The segmented structures included one inlet at the CHA, and 23 and 46 outlets for subject 1 and 2, respectively. The inlet diameter was 3.5 and 4.7 mm, respectively. These values are in agreement with reported CHA diameter ranges [3]. The smallest detected outlet diameter was ~0.5 mm, indicating the ability of our method to segment small vessels. The outlet generation is also shown with different colors such that CHA was considered as the first generation and the generation number was increased by one after each bifurcation. Results showed that the proposed pipeline can be successfully applied to segment the HA tree in different anatomies allowing for more efficient dosimetry compared to current clinical methods. Conclusions: We proposed a pipeline for segmentation of HA tree from CBCT images as the first step of developing a new patient-specific dosimetry protocol for liver cancer radioembolization planning. This new approach is based on CFD simulation and Y-90 radioactive decay physics. Our proposed pipeline provides the geometry (i.e. fluid domain of interest) needed for the CFD simulation for each patient. Future works will focus on predicting the distribution of the radioactive microspheres in the HA branches through CFD simulation, which in turn can help developing a more patient-specific dosimetry approach. Research Support: NIH R35CA197608 and CCSG P30 (NCI P30CA093373). ER -