Automatic localization of the left ventricle in cardiac MRI images using deep learning

Automatic localization of the left ventricle (LV) in cardiac MRI images is an essential step for automatic segmentation, functional analysis, and content based retrieval of cardiac images. In this paper, we introduce a new approach based on deep Convolutional Neural Network (CNN) to localize the LV in cardiac MRI in short axis views. A six-layer CNN with different kernel sizes was employed for feature extraction, followed by Softmax fully connected layer for classification. The pyramids of scales analysis was introduced in order to take account of the different sizes of the heart. A publically-available database of 33 patients was used for learning and testing. The proposed method was able it localize the LV with 98.66%, 83.91% and 99.07% for accuracy, sensitivity and specificity respectively.