چكيده به لاتين
Cardiovascular disease is the most common cause of death in the world. The mitral valve is one of the heart valves that are at the risk of disease. Echocardiography is used as an initial diagnostic tool for the evaluation of patients with heart valve disease. In this study, the diagnosis of mitral valve stenosis (MS) from two-dimensional echocardiography was investigated. The direct method of obtaining an area of the mitral valve is a common method for diagnosing MS. In this method, the echocardiologist performs measurements by manually tracking the mitral valve orifice in the mid-diastole frame. Since manual methods are time-dependent and user-dependent, in this study a new method is proposed for determining the mid-diastole frame automatically. The automatic segmentation of this frame is presented in the next section. Two different methods have been used to determine the mid-diastole frame. In the first method, by calculating the mean intensity within the ROI region, the mid-diastole frame is determined. In the second method, using the Principal Component Analysis algorithm (PCA) on the mitral valve orifice, in each frame of the echocardiography, mid-diastole frame was obtained, in which the high-dimensional ROI is mapped to a point in the two-dimensional space (each point represents a frame). A curve is obtained by calculating the Euclidean distance between consecutive points. By analyzing the curve, the mid-diastole frame is determined. The mean error in the determination of the mid-diastole frame for subjects with mitral valve stenosis and without MS for the first method was 9.36 and 2.5 respectively and for the second method was 0.72 and 1.87, respectively. For segmentation, three different methods based on the intensity have been used (including K-means, fuzzy C-means, and thresholding). In another section of this study, machine learning methods are used to diagnose the mitral valve stenosis. The best results were obtained using the SIFT features and accuracy for three different classifiers including SVM, ELM and Adaboost was 93/93, 97.23 and 93.83 respectively.
