High quality real-time image-to-mesh conversion for finite element simulations

Abstract

In this paper, we present a parallel Image-to-Mesh Conversion (I2M) algorithm with quality and fidelity guarantees achieved by dynamic point insertions and removals. Starting directly from an image, it is able to recover the isosurface and mesh the volume with tetrahedra of good shape. Our tightly-coupled shared-memory parallel speculative execution paradigm employs carefully designed contention managers, load balancing, synchronization and optimizations schemes which boost the parallel efficiency with little overhead: our single-threaded performance is faster than CGAL, the state of the art sequential mesh generation software we are aware of. The effectiveness of our method is shown on Blacklight, the Pittsburgh Supercomputing Center's cache-coherent NUMA machine, via a series of case studies justifying our choices. We observe a more than 82% strong scaling efficiency for up to 64 cores, and a more than 95% weak scaling efficiency for up to 144 cores, reaching a rate of 14.7 Million Elements per second. To the best of our knowledge, this is the fastest and most scalable 3D Delaunay refinement algorithm.

Fine-grained Speculative Topological Transformation Scheme for Local Reconnection Methods

“…To the best of our knowledge, this is the first non-Delaunay fine-grain tightly-coupled parallel method that optimizes the mesh connectivity in parallel throughout the generation procedure, including a post-improvement step. Similar tightly-coupled parallel Delaunay-based methods using speculative execution model appeared in [18,53]. The connectivity optimization is based on a speculative approach that has been proven to perform well on hardware-shared memory (i.e., single chip).…”

Toward a real time multi-tissue Adaptive Physics-Based Non-Rigid Registration framework for brain tumor resection

“…This is neither physically acceptable nor numerically feasible, indicating the need for choosing an adaptive FE biomechanical model to estimate the brain deformations, as is applicable to our case. We deployed, for this reason, a robust Delaunay mesh generation technique (Foteinos and Chrisochoides, 2013), which facilitates the FE model adaptivity, thereby creating high quality tetrahedral elements in parallel from a gradually warped pre-op segmented MRI. Rather than a local mesh generation, a global mesh generation is preferable because the introduced overhead is negligible in the former compared to the geometric complexity of the latter.…”

“…Parallel Block Matching (Liu et al, 2014) computes the displacements between the selected feature blocks and blocks in the intra-op MRI. The Parallel Mesh Generation (Foteinos and Chrisochoides, 2013) creates a two-tissue tetrahedral mesh from the warped pre-op segmented MRI. The FEM Solver (Clatz et al, 2005; Liu et al, 2014) builds a FE biomechanical model, applies the block matching displacements to the model, and estimates the mesh deformations iteratively by first using an approximation method.…”

“…The parallel mesh generation module is based on the Delaunay meshing and refinement algorithm developed in our group by Foteinos and Chrisochoides (2013). Let us denote the mesh in the beginning of an APBNRR iteration i with M i .…”