Efficient Convolutional Neural Network Model to Segment Left Ventricle from MRI Images

Abstract

Cardiovascular diseases can be avoided from being worse through early diagnosis by automatic segmentation of the left ventricle (LV). The LV is the most important chamber among the four heart chambers to diagnose cardiovascular disease due to its capability of pumping oxygenated blood to all parts of the body. Thus, segmentation of the LV from cardiac magnetic resonance imaging (MRI) is an essential step to obtain full morphological structures of LV and quantify global and regional cardiac function. However, the segmentation of the LV remains challenging due to the complex structure of MRI and various changes in the LV shape caused by different cardiovascular diseases.  To address this issue, a convolutional neural network (CNN) model was proposed to segment the LV from short-axis MRI images. The model was trained end-to-end from input images and their corresponding ground truths to classify each pixel in the images to segment the LV contours and myocardium. Training and testing phases were carried out by fine-tuning the network’s hyper-parameters with a learning rate of (0.01) and stochastic gradient descent (SGD) algorithm to achieve optimal performances. The proposed model was evaluated using metrics such as global and mean accuracy, mean and weight of intersection over union (IoU), and mean boundary F1 (BF) score. Results show the robustness of the proposed model with a rigid capability to segment the LV contours which is applicable for doctors to diagnose cardiac diseases at early stages with less effort and consumed time.