Model-based, knowledge-based epicardial boundary detector

A method was developed to detect the ventricle wall for the parasternal short axis view in two-dimensional echocardiography. The ventricular shape is modeled with a very simple parametric closed curve: an ellipse defined by four parameters. A genetic algorithm (GA) guides the elliptical shape and its position to find an optimum result. It minimizes one border dependent function, obtained after sampling 56 points in the ellipse. These points are located at the septum region (19) and at the ventricular posterior wall (37). The obtained solution is considered excellent as a first approximation to the real border. Then a second search complements the final solution. The ventricular shape is followed along one image sequence, basing the search space on the solutions given by the previous image. Medical knowledge about the cardiac structure and the image acquisition process is added to evaluate the function.<
>