Algorithmic Generation of Parameterized Geometric Models of the Aortic Valve and Left Ventricle.

Simulating the cardiac valves is one of the most complex tasks in cardiovascular modeling. As fluid-structure interaction simulations are highly computationally demanding, machine-learning techniques can be considered a good alternative. Nevertheless, it is necessary to design many aortic valve geometries to generate a training set. A method for the design of a synthetic database of geometric models is presented in this study. We suggest using synthetic geometries that enable the development of several aortic valve and left ventricular models in a range of sizes and shapes. In particular, we developed 22 variations of left ventricular geometries, including one original model, seven models with varying wall thicknesses, seven models with varying heights, and seven models with varying shapes. To guarantee anatomical accuracy and physiologically acceptable fluid volumes, these models were verified using actual patient data. Numerical simulations of left ventricle contraction and aortic valve leaflet opening/closing were performed to evaluate the electro-physiological potential distribution in the left ventricle and wall shear stress distribution in aortic valve leaflets. The proposed synthetic database aims to increase the predictive power of machine-learning models in cardiovascular research and, eventually, improve patient outcomes after aortic valve surgery.