Left ventricular contour detection: a fully automated approach

摘要

An automated approach for the detection of the left ventricular outline in a selected frame is described. The system uses a combination of algorithms to detect a rough model of the left ventricular contour. The first algorithm, a pyramidal segmentation algorithm, segments the image into a relatively large number of homogeneous areas. The second algorithm uses a neural network to select the areas that most likely belong to the left ventricle. After model detection, the actual contour is found by dynamic programming. The position of the aortic valve is assessed from the shape of the detected contour.