Transcatheter arterial chemo-embolization (TACE) is a therapeutic procedure to treat primary and metastatic liver cancer. It requires prior delineation of the hepatic arteries on magnetic resonance angiography (MRA) data and identification of the vessels supplying the tumor. Manual segmentation is extremely challenging and time consuming, thereby increasing the risk of wrongfully identifying the feeding vessels. We present a vascular path planning tool for TACE procedures by automatically segmenting the hepatic arteries on MRA. The proposed method first detects the celiac trunk from the aorta, then localizes and tags bifurcations throughout the arterial network for path planning. The algorithm is based on a multiple hypothesis tracking approach used to propagate deformable mesh surfaces. We validated the proposed framework on 20 liver-cancer-patients using abdominal MRA with 20 seconds delay after contrast injection. We show that the algorithm improves the selectivity of the arterial segments and outperforms two state-of-the-art methods with respect to manual segmentation, yielding mesh geometries with a mean absolute distance of 0.61 ± 0.10 mm for 3D mesh models.