Development and validation of a personalizable model of the hepatic arterial tree and particle transport
Embargo until
2020-08-01
Date
2019-04-25
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
Liver cancers are increasingly common and require improved treatment methods. Radioembolization inhibits tumor growth by inserting small spheres into the hepatic artery that are carried into smaller vessels near tumors, where they deposit radiation. Predicting toxicity for this treatment is challenging due to the microanatomical liver structure and the tendency of microspheres to cluster when trapped.
This work describes and validates a novel method to generate personalized hepatic arterial tree models and simulate realistic glass microsphere distributions. It describes methods to generate realistic arterial trees using physically constrained macrocell growth models that can incorporate measurable parameters such as initial and terminal vessel diameters and blood pressure. It also describes models of microsphere infusion that simulate different embolic effects and trapping behavior. Microsphere cluster-size histograms and images at PET resolution were simulated for several infusion methods in a hepatic arterial tree model generated from liver and proper hepatic artery segmentations taken from the XCAT phantom.
To verify the distributions created through microsphere infusion simulation, a characterization of the three-dimensional clustering of glass microspheres in tissue was necessary. However, traditional methods to find microsphere locations requires creating thin slices of treated tissue that are stained and examined microscopically, which is a very time-consuming process. This thesis describes how to use micro-CT and image processing to detect glass microspheres in whole tissue samples. The detection method was verified by processing phantoms composed of glass microspheres embedded in agar that could also be imaged microscopically to determine true microsphere locations. Using this, >93% of microspheres in the phantom were correctly detected and the percentage of erroneously detected microspheres was <5%. The mean absolute error between the true and estimated dose maps was 4.2%.
The detection method was then applied to a treated porcine liver sample to characterize microsphere clustering behavior in normal liver tissue. A histogram of these cluster sizes was compared with histograms simulated using the tree and transport models. The most realistic clustering behavior was produced by having microspheres trap in vessels when their diameters exceeded the vessel diameters, and when embolic effects did not affect later microsphere traversal.
Description
Keywords
Radioembolization, Microspheres, Microdosimetry, Treatment planning