The health of the heart is a crucial issue in human’s health and the cardiovascular disease can be diagnoses by several imaging methods. Magnetic resonance is one of the popular methods in cardiovascular imaging and the segmentation of left ventricle is important for diagnosing such diseases in magnetic resonance images. There are challenges for this task, including the similarity of shape and intensity of left ventricle comparing other organs in the image, inaccurate boundaries and presence of noise in most of the images. There exist numerous segmentation methods which could be categorized based on using the prior information. In chapter three we propose a semi-automated segmentation method which tracks the left ventricle in a cardiac cycle. It works based on a proposed cost function and search area. The advantage of this method is its independency of any training phase and has appropriate output for most of cardiac magnetic resonance images in temporal space. We propose an automated method for segmenting the left ventricle in cardiac MR images. First, we automatically extract the region of interest and then employ it as an input to a fully convolutional network. We accurately trained the network despite the small amount of training data and also left ventricle pixels in comparison with the whole image. A thresholding is applied on the output map of the fully convolutional network and selection of regions based on their roundness is performed in our proposed post-processing phase. These methods are evaluated by dice metric and achieve to 86.84% for the first method and 87.24% for the second method on the heart dataset of York University. Keywords: Segmentation, left ventricle segmentation, MRI segmentation, LV segmentation in MRI, block matching, tracking, deep learning
