Virtual parenchymal perfusion for selective intra-arterial therapy of liver cancer

Intra-arterial injection of therapeutic agents is among the most common options for unresectable liver cancer. Such procedures rely on an in-depth assessment of the vessels leading to the tumor(s) to select the optimal catheter location maximizing therapy selectivity and potency while limiting healthy liver toxicity. In this paper, we propose a Virtual parenchymal Perfusion (VP) algorithm to approximate distal liver vascular territories given prospective "virtual" injection points, with all predictive data derived from a single 3D arterial Cone Beam Computed Tomography (CBCT) scan. Vascular territories are extracted from a vessel-driven geodesic distance map computed from the user-selected injection position, without the need for explicit vessel segmentation. In order to evaluate our approach, we compared liver sub-territories segmented from actual CBCT scans after selective injection of contrast agent or lipiodol, with the vascular territories computed with the VP algorithm using the non-selective arterial CBCT scans. Results on 11 pairs of arterial and selective injection CBCT scans showed a non-significant difference between virtual and real segmented volumes (p=0.73), an average mean distance of 4.75 mm, an average volume error of 16% and an average Dice coefficient of 0.75, supporting a high correlation. This study demonstrates the potential of our VP algorithm in planning intra-arterial therapy of liver cancer and provides the basis for clinical translation in a live interventional context.