Automatic Left Ventricle Segmentation in Cardiac Magnetic Resonance Images via Threshold Selection

Abstract

In medical diagnosis, the movement of the myocardium of left ventricle (LV) can represent the pump function of the heart, which can provide the basis for diagnosis of heart diseases. Magnetic resonance imaging (MRI) is an effective tool for the clinical diagnosis of heart diseases due to its special imaging mechanism, which is particularly effective for soft tissue such as heart. Identification of the LV endocardium, especially the apical and basal slice images and some special mid-ventricular slice images with poor image quality, is still a very challenging and open problem. In this paper, an automatic segmentation method based on threshold is proposed. This method works well in some mid-ventricular slices with poor image quality and some ventricle slices with messy edge. We tested the proposed method with other 15 popular segmentation algorithms by ten frames of Cardiac MRI at end systole (ED) or end diastole (ES) phases. Those frames of images are difficult to segment the LV endocardium from Medical Image Computing and Computer Assisted Intervention (MICCAI) 2009 challenge. Those frames of images are from apical and basal slices or mid-ventricular slices that are difficult to segment the LV endocardium. Finally, we assessed the deviation between the automatically segmented and benchmark manual contours. The proposed method achieved 0.9384 average Dice metric, 1.2715 mm average perpendicular distance (APD). These results compared with other algorithms demonstrate that the proposed method is an effective and viable method to identify the LV endocardium at ED and ES phases.