Improving ejection fraction estimation for 2D ultrasound using a computer-generated cardiac model

As a tool to develop a better ejection fraction estimate, we have adapted a cardiac torso computer-generated model that closely resembles the anatomical structures and simulates cardiac and respiratory motions of a normal human subject. Prior work includes a dynamic Non-uniform Rational B-spline (NURBS)-based cardiac torso (NCAT) phantom based on CT and MRI clinical data [1–3]. Because the valves have not been modeled yet in the current version of the NCAT software, we used activity phantoms and assigned different activity parameters for the left atrium and the left ventricle, so that they could be distinguished. In order to mimic standard ultrasound (US) cardiac views, NCAT 2D slices were selected and then manipulated by contrast enhancement, resolution improvement, and negative imaging. These results indicate that the 4D simulation model is a more accurate estimator of the left ventricular volume and ejection fraction than the earlier 2D methods. 4D US clinical data is being obtained to independently verify the model. Scan planes from the model were matched to actual 2D US cardiac image loops through scaling and feature correspondence. Challenges include absence of certain clinical landmarks such as valves in the model and wall dropouts in the images. The adapted NCAT phantoms can be used for US imaging demonstrations and the simulation of cardiac disease.