Near Real-time Adaptive Isotropic and Anisotropic Image-to-mesh Conversion for Numerical Simulations Involving Cerebral Aneurysms

TL;DR

The paper tackles the challenge of real-time discretization of complex cerebral vasculature for CFD by presenting two complementary I2M workflows. The first constructs adaptive anisotropic meshes via a pipeline that merges CBC3D, AFLR, and CDT3D on cc-NUMA architectures, enhanced by hierarchical load balancing and an optimized local reconnection; it achieves near real-time performance for million- to tens-of-m millions-element meshes. The second extends PODM to produce adaptive isotropic meshes using a user-defined sizing function, enabling rapid generation of large isotropic meshes. Together, they demonstrate feasible near real-time I2M generation for aneurysm simulations on high-end multicore systems, with detailed analyses of quality, fidelity, and scalability, and lay out a roadmap for full integration into clinical simulation pipelines.

Abstract

Presented are two techniques that are designed to help streamline the discretization of complex vascular geometries within the numerical modeling process. The first method integrates multiple software tools into a single pipeline which can generate adaptive anisotropic meshes from segmented medical images. The pipeline is shown to satisfy quality, fidelity, smoothness, and robustness requirements while providing near real-time performance for medical image-to-mesh conversion. The second method approximates a user-defined sizing function to generate adaptive isotropic meshes of good quality and fidelity in real-time. Tested with two brain aneurysm cases and utilizing up to 96 CPU cores within a single, multicore node on Purdue University's Anvil supercomputer, the parallel adaptive anisotropic meshing method utilizes a hierarchical load balancing model (designed for large, cc-NUMA shared memory architectures) and contains an optimized local reconnection operation that performs three times faster than its original implementation from previous studies. The adaptive isotropic method is shown to generate a mesh of up to approximately 50 million elements in less than a minute while the adaptive anisotropic method is shown to generate approximately the same number of elements in about 5 minutes.

Figures (15)

• Figure 1: Presented is the pipeline of software tools used to satisfy the medical image-to-mesh conversion requirements for use within a medical numerical simulation.
• Figure 2: The original image of the second (middle cerebral artery bifurcation) aneurysm case is shown from an anteroposterior view. (a) shows the vascular structure imaged by rotational angiography of the right carotid artery. (b) zooms in to the region of interest while (c) highlights the PLC obtained from the segmentation of the aneurysm. (d) shows only the resulting PLC.
• Figure 3: Shown is streamline data of the velocity field from a CFD simulation with the second (middle cerebral artery bifurcation) aneurysm case. (a) and (b) show different viewpoints. The velocity color legend is in m/s CFDDrivenTopology2024.
• Figure 4: Cross sections are shown of the volume meshes generated by CBC3D and TetGen for the first aneurysm case, merged with the conforming boundary layer mesh generated by AFLR.
• Figure 5: Cross sections are shown of the volume meshes generated by PODM and CDT3D for the first aneurysm case, where the CDT3D meshes are merged with the conforming boundary layer mesh generated by AFLR.