Overview of Left Ventricular Segmentation in Ultrasound Images

Abstract

Due to its great temporal resolution and quick acquisition periods, two-dimensional echocardiography, or shorter 2D echo, is the most used non-invasive approach for assessing heart disease. It offers a grayscale image that anatomical details can be extracted from to evaluate heart functioning. The initial stage in quantifying cardiac function in 2D echo is the segmentation of the left ventricular (LV) walls. The primary boundary identification methods used for 2D echo at the moment are semi-automatic or manual delineation carried out by professionals. However, manual or semi-automatic approaches take a lot of time and are subjective, which makes them vulnerable to both intra- and inter-observer variability. Many researchers have tried to automate the process of left ventricle segmentation. The extensive use of deep learning algorithms has lately changed medical image analysis. The revolution has primarily been powered by supervised machine learning with convolutional neural networks. In this paper, we will provide a short overview of some of the popular deep-learning techniques for left ventricular segmentation in two-dimensional echocardiography.