Left ventricle segmentation by combining convolution neural network with active contour model and tensor voting in short-axis MRI

Abstract

Left ventricle(LV) segmentation is a prerequisite step of evaluation of LV structure and function, which plays an important role in the diagnosis and treatment of cardiovascular diseases. In this paper, we propose a method to segment endocardium and epicardium of LV using convolution neural network combined with active contour model and tensor voting. A fully convolution neural network (FCN) named VGG16 is employed to segment myocardium of LV firstly. To improve the segmentation accuracy of endocardium, active contour model is employed to segment endocardium based on the initial segmentation results of FCN. Furthermore, to deal with the discontinuity of epicardium, tensor voting is used to fill the missing parts of myocardium. Finally, ellipse detection is employed to prune surplus parts in epicardium. Experiments on public datasets demonstrate that our method outperform most existed automated segmentation method in respect of several commonly used evaluation measures.