Automating Model Generation for Image-Based Cardiac Flow Simulation

Abstract

Computational fluid dynamics (CFD) modeling of left ventricle (LV) flow combined with patient medical imaging data has shown great potential in obtaining patient-specific hemodynamics information for functional assessment of the heart. A typical model construction pipeline usually starts with segmentation of the LV by manual delineation followed by mesh generation and registration techniques using separate software tools. However, such approaches usually require significant time and human efforts in the model generation pro-cess, limiting large-scale analysis. In this study, we propose an approach towards fully automating the model generation process for CFD simulation of LV flow to significantly reduce LV CFD model generation time. Our modeling framework leverages a novel combination of techniques including deep-learning based segmentation, geometry processing, and image registration to reliably reconstruct CFD-suitable LV models with little-to-no user intervention. We utilized an ensemble of 2D convolutional neural networks (CNNs) for automatic segmentation of cardiac structures from 3D patient images and our segmentation approach outperformed recent state-of-the-art segmentation techniques when evaluated on benchmark data containing both MR and CT cardiac scans. We demonstrate that through a combination of segmentation and geometry processing, we were able to robustly create CFD-suitable LV meshes from segmentations for 78 out of 80 test cases. Although the focus on this study is on image-to-mesh generation, we demonstrate the feasibility of this framework in supporting LV hemodynamics modeling by performing CFD simulations from two representative time-resolved patient-specific image data sets.

Whole Heart Mesh Generation For Image-Based Computational Simulations By Learning Free-From Deformations

“…Our supplementary materials additionally include an ablation study of individual loss components and the use of GCN decoder. Furthermore, we compared these FFD-based methods with prior whole-heart reconstruction or segmentation methods, Kong et al [10], 2DUNet [17,9], residual 3D UNet [7] and Voxel2Mesh [23]. We followed procedures described in [10] to implement those methods.…”

“…However, generating simulationsuitable models of the heart from image data requires significant time and human efforts and is a critical bottleneck limiting clinical applications or large-cohort studies [13,14]. Deforming-domain computational fluid dynamics (CFD) simulations of the intracardiac hemodynamics, in particular, requires both the geometries and the deformation of the heart from a sequence of image snapshots of the heart throughout the cardiac cycle [13,21,9]. Challenges of image-based model construction are related to the entwined nature of the heart, difficulty differentiating individual cardiac structures from each other and the surrounding tissue, the large deformations of these structures over the cardiac cycle, as well as complicated steps to label various surfaces or regions for the assignment of boundary conditions or parameters if the model is to be used to support simulations.…”

“…However, prior studies have not generally focused on learning to predict surface meshes directly from medical data for the purpose of computational simulations. Prior efforts on accelerating cardiac model construction for simulations have typically adopted a multistage approach whereby 3D segmentations of cardiac structures are first obtained from image volumes using convolutional neural networks (CNN), meshes of the segmented regions are then generated using marching cube algorithms, and finally manual surface post-processing or editing is performed [9,11]. However, CNN based methods may produce segmentations that achieve high average voxel-wise accuracy but contain extraneous regions that are anatomically inconsistent, unphysical and unintelligible for simulation-based analyses.…”

“…However, CNN based methods may produce segmentations that achieve high average voxel-wise accuracy but contain extraneous regions that are anatomically inconsistent, unphysical and unintelligible for simulation-based analyses. Correcting such artifacts generally requires a number of carefully designed post-processing steps [9] that cannot be easily adapted to generating complicated whole heart models.…”

A Deep-Learning Approach For Direct Whole-Heart Mesh Reconstruction

“…Prior cardiac model construction efforts have typically adopted a multistage approach whereby 3D segmentations of cardiac structures are first obtained from image volumes, meshes of the segmented regions are then generated using marching cube algorithms, and finally manual surface post-processing or editing is performed (Lorensen and Cline, 1987;Kong and Shadden, 2020;Maher et al, 2019;Augustin et al, 2016). The quality of reconstructed surfaces highly depends on the quality of segmentation and the complexity of the anatomical structures.…”