Transforming body composition of computational phantoms based on form-finding dynamics.

BACKGROUND: There is a lack of representation available for computational phantoms with variable muscle and fat composition representing the overall population. Changes in body composition are known to significantly affect image quality of patients. Therefore, it is critical to capture this variability to conduct virtual imaging trials representative of the full spectrum of patients. Current methods require skilled 3D artists or finite element modeling, which are both time and resource intensive. A potential alternative is the use of a physics engine, which balances computational speed and realism. PURPOSE: The purpose of this study is to model changes in muscle and fat composition in the computational XCAT phantom library to simulate additional anatomic variations due to visceral fat (VF), subcutaneous fat (SF), and skeletal muscle (SM) growth or reduction. The modeling considers the effects of the changes on surrounding organs. METHODS: Algorithmic modeling of the changes was done through the live physics engine Kangaroo2. Kangaroo2 is mainly used to create architecture using its physics driven form-finding optimization or simulate physical interactions between structures. Muscle and other soft tissue were first initialized by converting phantom image datasets into isotropic triangular meshes for accurate simulation. Each element within the respective meshes was then defined by various conditions, such as pressure or stiffness. All conditions were added to the Kangaroo engine solver, and the overall response of the system was calculated. Following simulations, a diffeomorphic displacement field was calculated based on the voxelized initial and final mesh. The large deformations were applied to a high resolution image of the original phantom to get the final result. We applied the model with initial settings to a test phantom and modeled three different changes: VF, SF, and SM growth. We also applied the model to an obese phantom to model VF and SF reduction. RESULTS: The techniques that we devised successfully changed the volumes and appearances of the different components within the phantoms along with local organ deformation. Total volume and cross sectional area at the L3 slice were measured for each phantom change. For the growth phantoms, the volume growth of the main region of interest was measured to be greater by a factor of 12.9, 7.3, and 1.6. The reduction algorithm reduced the SF volume by a factor of 0.63. Reference z-scores of each component were then calculated to show the new position of the phantom within a real population. CONCLUSIONS: The techniques that we developed allow us to manipulate different phantoms by changing their muscle and fat composition. Changes can be observed qualitatively and quantitatively. This workflow can become useful to expand existing phantom libraries for better representation of body composition variability within observed patient populations.

Duke Scholars

Author William Paul Segars Radiology