Automatic bone removal for 3D liver tace planning with C-arm CT

Transarterial chemoembolization (TACE) involves a targeted catheter-based injection of a chemotherapeutic and an embolic agent into the feeding vessels of a liver tumor. For this task, contrast-enhanced C-arm computed tomography (C-arm CT) is gaining popularity, as it offers three-dimensional (3D) imaging of vessels and tumors during the intervention. The identification and segmentation of tumor feeding vessels is preferably performed using volume visualization methods. However, the presence of surrounding bone structures impedes un-obstructed vessel visualization and may impact vessel segmentation. We present an approach for fully automated bone removal to facilitate 3D TACE planning. A model of the `visceral cavity´, containing the internal organs, is adapted to the target image; bones outside the adapted model are removed. The method is evaluated on 30 clinical cases. In a quantitative evaluation against manual segmentations, the mean mesh-to-mesh distance is 2.7 mm, with best performance in the spine area (1.1 mm) and worst in the anterior portion of the C-arm CT volumes (4.4 mm). A visual inspection of volume renderings shows satisfactory bone removal in 97% of cases, 70% of which show complete bone removal.