Optimization of an edge detection algorithm for echocardiographic images

Echocardiography is currently the predominant technique to assess cardiac function. The majority of quantitative diagnostic parameters require the calculation of ventricle volumes computed from two-dimensional pictures. Consequently, the segmentation of heart cavity remains a crucial step. We present the optimization method of a nearly automated edge detection algorithm applicable to each kind of cross section. This algorithm is based on a Cohen's Balloon deformable model improved by providing an initial guess that is very close to the contour. The first step consists of a preprocessing thresholding and a morphological opening. The second step provides the initial guess, a B-spline built from a radial sampling within a cavity, whose centre has previously been pinpointed by the operator. During the last step, the balloon based algorithm tracks towards the best border solution. The two first stages of our optimization process led to the design of the previously described algorithm. The results of the third stage are presented in this paper. We have optimized the parameters involved in the detection of the initial edge condition by including prior knowledge. We assessed the algorithm with a set of twenty eight images. The first results are very encouraging and will be improved during the fourth stage: the optimization of the parameters of the snake.