Abstract

Info

Title:: Unsupervised image decomposition of DCE-MRI based on generative adversarial networks

Author(s):: MI Jia; ZHOU Yujia; FENG Qianjin; School of Biomedical Engineering, Southern Medical University/Key Laboratory for Medical Image Processing in Guangdong Province, Guangzhou 510515, China

Keywords:: Keywords: deep leaning generative adversarial network image decomposition dynamic contrast-enhanced magnetic resonance imaging

Abstract:: Abstract: Image decomposition can be used to separate the grayscale variations caused by contrast agent from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) image, with other contents unchanged, thereby eliminating the effects of contrast agent on quantitative analysis. An unsupervised image decomposition method based on generative adversarial networks is proposed to achieve the separation of grayscale variations caused by contrast agent in DCE-MRI. The method assumes that the enhanced image is composed of contrast component and anatomy component, and two generators are designed based on the complexity of each component. Through the double-generator, the information of anatomy component and contrast component can be extracted simultaneously and independently. Meanwhile, the image reconstruction loss is introduced as a feedback signal to constrain the generation. Mean-square error, peak signal to noise ratio and structural similarity index are used to compare the proposed method with the existing methods on MNIST dataset, thus verifying the effectiveness of the proposed method. Moreover, some experiments are carried out on clinical liver DCE-MRI data, and the experimental results show that the proposed method can be used to effectively achieve DCE-MRI image decomposition without paired training data.