Abstract
Microbubble enhanced High Intensity Focused Ultrasound (HIFU) is of great interest to tissue ablation for solid tumor treatments such as in liver and brain cancers, in which contrast agents/microbubbles are injected into the targeted region to promote heating and reduce pre-focal tissue damage. A compressible Euler-Lagrange coupled model has been developed to accurately characterize the acoustic and thermal fields during this process. This employs a compressible Navier-Stokes solver for the ultrasound acoustic field and a discrete singularities model for bubble dynamics.
To address the demanding computational cost in practical biological applications, a multi-level hybrid MPI-OpenMP parallelization scheme is developed to take advantage of both scalability of MPI and load balancing of OpenMP. At the first level, the Eulerian computational domain is divided into multiple subdomains and the bubbles are subdivided in groups based on which subdomain they fall into. At the next level, in each subdomain containing bubbles, multiple OpenMP threads are activated to speed up the bubble computations. More OpenMP threads are used inside each subdomain where the bubbles are clustered. By doing this, MPI load imbalance issue due to non-uniformity of bubble presence is compensated.
The hybrid MPI-OpenMP Euler-Lagrange solver is used to conduct simulations and physical studies of bubble-enhanced HIFU problems containing a large number of microbubbles. The phenomenon of acoustic shadowing caused by the bubble cloud is then analyzed and discussed. Hybrid parallelization efficiency tests and demonstration of its advantages against using MPI alone are presented.
