Anatomically and physiologically informed computational model of hepatic contrast perfusion for virtual imaging trials

Abstract

Purpose Virtual (in silico) imaging trials (VITs), involving computerized phantoms and models of the imaging process, provide a modern alternative to clinical imaging trials. VITs are faster, safer, and enable otherwise‐impossible investigations. Current phantoms used in VITs are limited in their ability to model functional behavior such as contrast perfusion which is an important determinant of dose and image quality in CT imaging. In our prior work with the XCAT computational phantoms, we determined and modeled inter‐organ (organ to organ) intravenous contrast concentration as a function of time from injection. However, intra‐organ concentration, heterogeneous distribution within a given organ, was not pursued. We extend our methods in this work to model intra‐organ concentration within the XCAT phantom with a specific focus on the liver. Methods Intra‐organ contrast perfusion depends on the organ's vessel network. We modeled the intricate vascular structures of the liver, informed by empirical and theoretical observations of anatomy and physiology. The developed vessel generation algorithm modeled a dual‐input‐single‐output vascular network as a series of bifurcating surfaces to optimally deliver flow within the bounding surface of a given XCAT liver. Using this network, contrast perfusion was simulated within voxelized versions of the phantom by using knowledge of the blood velocities in each vascular structure, vessel diameters and length, and the time since the contrast entered the hepatic artery. The utility of the enhanced phantom was demonstrated through a simulation study with the phantom voxelized prior to CT simulation with the relevant liver vasculature prepared to represent blood and iodinated contrast media. The spatial extent of the blood–contrast mixture was compared to clinical data. Results The vascular structures of the liver were generated with size and orientation which resulted in minimal energy expenditure required to maintain blood flow. Intravenous contrast was simulated as having known concentration and known total volume in the liver as calibrated from time–concentration curves. Measurements of simulated CT ROIs were found to agree with clinically observed values of early arterial phase contrast enhancement of the parenchyma (∼5$ \sim 5$ HU). Similarly, early enhancement in the hepatic artery was found to agree with average clinical enhancement false(180$(180$ HU). Conclusions The computational methods presented here furthered the development of the XCAT phantoms allowing for multi‐timepoint contrast perfusion simulations, enabling more anthropomorphic virtual clinical trials intended for optimization of current clinical imaging technologies and applications.

Development and application of a virtual imaging trial framework for airway quantifications via CT

“…Twelve anthropomorphic, computational 4D extended cardiac-torso (XCAT) phantoms consisting of six female and six male body habitus with varying body mass indices (BMIs) were developed with varied severity of bronchitis. Similar to how the original XCAT phantoms were developed [8][9][10][11][12][13], the overall body habitus of these phantoms was created by segmenting the major organs [12]. The anatomy of each phantom included highly detailed, mathematically extended models of the airways and vessels [10].…”

In silico simulation of hepatic arteries: An open‐source algorithm for efficient synthetic data generation

“…Previously proposed methods for the simulation of 2D x-ray images include, DeepDRR, 32 a convolutional neural network which transforms an input image of a forward projected CT volume into a realistic 2D x-ray image.However,DeepDRR was developed for noncontrast enhanced CT volumes and thus the resulting images do not contain visible vasculature. Algorithms for the formation of synthetic vasculature, which may be included in existing CHPs include fractal-based and diffusion-limited aggregation algorithms, [33][34][35][36] which use order-based geometrical constraints to automatically generate random vessel trees. These approaches can lack realism due to straight or repetitive segments.…”

A mesh-based model of liver vasculature: implications for improved radiation dosimetry to liver parenchyma for radiopharmaceuticals

“…[ 9 ] applied the CCO method in the construction of a hepatic arterial tree for stimulation studies of the infusion and trapping of Y-90 microspheres during hepatic tumor radioembolization. In 2022, Sauer et al similarly created a vascular network in the human liver for use in CT imaging studies of hepatic contrast perfusion [ 10 ]. Our work proposes a complete vascular network including the hepatic arterial, portal, and venous blood circulation in the livers of the ICRP reference adult male and adult female human computational phantoms [ 11 ], with applications for refined dose assessment of liver parenchyma for internal emitters.…”