Abstract No. 544 Evaluation of software predicting selective liver perfusion for embolization planning

Abstract

To evaluate the accuracy of a liver embolization planning software (Liver ASSIST V.I. with Virtual Parenchymography prototype, GE Healthcare, Chicago, IL) to automatically predict perfused liver volume from any virtual selective injection point on a proximally enhanced planning cone-beam CT (CBCT). Planning CBCTs acquired proximally and associated selective CBCT couples from patients undergoing liver embolization in two centers were included in this retrospective study. Inclusion criteria were as follows: perfused liver parenchyma volume on the selective CBCT > 100cc and with adequate enhancement to allow for manual contouring. Perfused liver parenchyma was contoured manually on all selective CBCTs, defining the ground truth volume. The Virtual Perfused Volume (VPV) was obtained with the software by selecting the virtual injection point on the proximal CBCT matching the actual injection point of the associated selective CBCT. Software accuracy was evaluated by comparing the VPV to the ground truth volume after registering the proximal and selective CBCTs based on liver borders. Software accuracy is reported as relative volumetric error (%) and Dice spatial overlap coefficient (%). Results are expressed as mean [interquartile range]. 32 CBCT couples were included in 18 patients. Proximal CBCTs were acquired from the common (52%), right (18%), left (9%) or segmental (21%) hepatic arteries. Selective CBCTs were acquired from the right (12%), left (19%), and third or greater order right (65%) or left (4%) hepatic artery, with an average perfused liver parenchyma volume of 512cc [192, 699]. Software was successful in automatically computing VPV in all cases. Relative volumetric error and Dice coefficient were 15% [5%-20%] and 77% [73%-84%] in the entire cohort. In the sub-analysis on the 27 couples with proximal CBCTs acquired from the common, right or left hepatic artery, relative volumetric error and Dice coefficient were 12% [4%-18%] and 78% [74%-84%], with no significant difference between centers (P = .31 and .13). Preliminary evaluation of selective perfused liver volume prediction from proximal CBCTs showed acceptable accuracy. This software could help identify optimal embolization strategy based on a single proximal CBCT, avoiding multiple selective acquisitions for liver embolization planning. Further studies are warranted to demonstrate its benefits for radioembolization CBCT-based volumetry, reducing the need for distal vessel catheterization during mapping, and streamlining dosimetry workflow.