Abstract

Biventricular statistical atlases are useful dimensionality reduction tools that have demonstrated utility for discovering improved diagnostic and prognostic markers. Patient-specific geometries used to generate these atlases are obtained from in-vivo tomographic images of end diastole (ED) and/or end-systole (ES). However, because these are pressure loaded states, variations in hemodynamic loading may impact downstream shape analysis. In this project, we used a computationally expensive, iterative method to estimate the ventricular geometry in the absence of hemodynamic loading and quantified the shape difference between pressure loaded and unloaded states using a statistical atlas. Next, we assessed whether these atlas-based shape differences could be used as an accurate initial guess for the iterative method, thereby improving the convergence time. In a cohort of 23 patients with chronic thromboembolic pulmonary hypertension (CTEPH), hemodynamic unloading produced atlas-based z-score changes that were highly conserved between patients, with absolute differences between the loaded and unloaded states of 1.0 ± 0.14, 1.2 ± 0.27, −0.1 ± 0.08, and 0.5 ± 0.08 for the first 4 shape modes, respectively. When we used these atlas-based z-score changes as an initial guess, we observed a 72% decrease in required CPU time to solve for patient-specific unloaded geometries.

This content is only available via PDF.
You do not currently have access to this content.