An implementation of a real-time finite element algorithm using CUDA technology with application to Endoclamp Balloon expansion

An Endoclamp Balloon achieves endovascular clamping by means of intraluminal occlusion. Circumventing the need for opening the thoracic cage provides significant benefits in terms of reducing trauma for the patient. This procedure, however, induces risks of damage to the aortic lumen by way of over-expansion of the balloon, exposing the aortic tissue to unsafe levels of strain. In patient-specific terms, accurate estimates of stress, strain and damage are required intraoperatively to warn the surgeon and mitigate the risk of tissue damage. Due to the nature of aortic tissue and the physical system in question, the Total Lagrangian Explicit Dynamic (TLED) finite element (FE) formulation [1], capable of handling geometric and material nonlinearities, is used. A 2D plane strain formulation is used to additionally simplify and speed up the execution time by reducing the number of degrees of freedom involved, meanwhile retaining sufficient accuracy for the proposed application. The large amount of computation needed to solve for the stress field is relieved by using dedicated massively parallel hardware. An Nvidia CUDA GPGPU(General Purpose Graphics Processing Unit) is employed to parallelize critical portions of the TLED algorithm. The fusion of sensory information provided by the Endoclamp Balloon and electromagnetic position sensors, along with a segmented patient-specific aortic mesh provide boundary conditions and localization information for the FE simulation. We present an efficient implementation of the TLED algorithm on a planar problem with 300 degrees of freedom and 1000 time steps solved within 65ms - thus complying with the time constraints imposed by the workflow of the surgical theater. The accuracy of the solution is validated against an industry-proven FE package Abaqus.

Publication year:2012

Accessibility:Closed