Reconstruction of Left Ventricle Geometry for Finite Element Analysis Using 2D Echocardiography

Patient-specific finite element (FE) analysis of left ventricular (LV) mechanics requires a geometric model that is discretized for computation, a step that in current practice still involves considerable manual work. Because two-dimensional (2D) echocardiography is the most widely used modality for assessing LV function, we present an algorithm that converts segmented apical 2D views (one to three imaging planes) directly into a structured, layered hexahedral FE mesh controlled by three LV-specific parameters: the axial and circular divisions and the number of transmural element layers. All models generated with the same parameters share one mesh topology, so node-to-node correspondence across cardiac phases is obtained by construction and tracked endocardial motion can be imposed directly as a boundary condition. On analytical phantoms with known 3D geometry, reconstruction from three apical views reproduced the cavity volume to within 2.1% (biplane Simpson method: up to 10.8%), with a mean surface distance of 0.6–0.9 mm at a pixel size of 0.8 mm and a Dice coefficient of 0.95–0.98; reconstruction from a single view underestimated the volume by 5–20%, depending on the ellipticity of the cross-section. Applied to 36 segmented EchoNet-Dynamic images, the pipeline produced valid meshes (all elements with a positive Jacobian) in every case and at every resolution, in less than 0.2 s per model. A generic marching-cubes/TetGen route applied to the same geometry produced different node counts at end-diastole and end-systole and contained sliver elements. In a passive-filling problem (human Holzapfel–Ogden myocardium with rule-based fibers, end-diastolic pressure 8 mmHg), cavity volume, wall thickness, apex displacement, fiber strain and fiber stress changed by at most 0.8% between the 24 × 48 and the 36 × 72 surface divisions. The method therefore provides fast, controllable and FE-ready LV geometries from routine echocardiography, with an accuracy that is limited mainly by the number of available views and by the segmentation.