Automatic Segmentation of the Left Ventricle Cavity from Cine MRI Images

The early diagnosis of cardiovascular diseases plays an important role in the recovery. Diagnosis can be done by evaluating cardiac function. Automatic left ventricle (LV) segmentation is an essential step in the evaluation of cardiac function. In this paper, a new method for segmenting the left ventricle cavity based on pixel classification is proposed. In the proposed method, a set of features (i.e. the gradient magnitude, the Laplacian of Gaussian (LOG) filter, and the pixel intensity value) are used. To show the local properties of neighbored pixels, a set of non-overlapped patches are used, followed by the principal component analysis (PCA) for feature reduction. Then the reduced feature vector with the ground truth labels are used with the K-nearest neighbor (KNN) classifier in the segmentation of the LV cavity. Experimental results illustrate the reliability of the proposed method which achieves sensitivity and specificity of 96.89 % and 98.7%.