| Two-dimensional real-time echocardiography is a noninvasive technique widely used in clinical cardiology. Although boundary detection is crucial to echocardiography, for it allows the computation of very important measures, automatic, even semi-automatic, boundary searching in echocardiograms is often difficult because of the poor quality of ultrasonic images. In this paper, a new algorithm called star algorithm is presented, which makes use of cost function driven system and special knowledge to detect endocardial boundary of left ventricle. Star algorithm was designed to overcome the problems of bad image quality and complex endocardial shape in echocardiographic endocardial boundary detection.Part I Basic experiment studyObjective In the basic experiment, the automatic endocardial boundary detection of star algorithm has been tested by apical four chamber view and mid-papilary short axis view images in the adapability of image quality and shape changes.The endocardial boundary detection of left ventricle in apical four chamber viewMethods The detecting strategy of star algorithm includes three steps, which are to locate the left ventricular chamber, to locate the chamber center of left ventricle and to detect the endocardial boundary from inside the chamber. According to the strategy, star algorithm is divided into following four stages.Image preprocess In star algorithm the image preprocess includes 3 operations, which are histogram equalization, contrast stretch and 5x5 neighborhood average smoothing. Left ventricular chamber location The only purpose of this stage is to find a point inside the left ventricular chamber. Therefore, eight radial lines are set out one by one to detect the endocardial boundary at 45°intervals around the image center. Each line has an edge detector in the front which is made of two parallel arcs with an eight-pixel distance. Each arc has eight sample points. As the edge detector goes along the radial line, it calculates the gray level gradient of the two arcs. The location of the maximal gradient point on each radial line is remembered as the protemporal boundary point of endocardial boundary and the detector goes back along the same line to find the opposite endocardial point along the line. The middle point of the two boundary points with the largest distance is taken as a point inside the left ventricular chamber. Left ventricular chamber center location From the detected point inside the left ventricular chamber, 60 radial scan lines are set out one by one at 6°intervals, each one having a different size edge detector in the front. The positions of maximum gradient points are remembered. The average coordinate of all the remembered points is taken as the position of the initial left ventricular chamber center. Then the process goes iteration until the position of the detected chamber center stop changing between iterations. Left ventricular endocardial boundary detection 60 radial scan lines are set out from the detected chamber center which remembers four maximum gradient points as candidates on each radial line. The maximum gradient point which is nearest the chamber center on each radial line is selected to be the initial boundary point. Then, the mean neighborhood distance of these points is calculated, of which a threshold is set empirically. The longest continuous boundary section, which means the neighborhood distances of all the points on the boundary section are below the threshold, is set to be the initial boundary section. Then, a cost function, using mean neighborhood distance and sectional mean gray level of boundary points, is applied to link the rest part of the boundary. After the boundary points are obtained, a B spline curve is used to obtain a smooth endocardial boundary.Results Star algorithm was performed on eight pairs of end-diastolic and end-systolic apical four chamber images which had various image quality problems. The result of the test turned to be that star algorithm worked soundly on these images. Compared wi... |
PDF Full Text Request