Unsupervised medical images denoising via graph attention dual adversarial network
- PDF / 9,091,792 Bytes
- 12 Pages / 595.224 x 790.955 pts Page_size
- 96 Downloads / 218 Views
Unsupervised medical images denoising via graph attention dual adversarial network Tianxu Lv1 · Xiang Pan1
· Yazhou Zhu1 · Lihua Li2
Accepted: 9 October 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract A lot of natural images denoising methods have been proposed, however, there are dual primary challenges for medical images denoising: 1) paired datasets are scarce and 2) medical images are often three-dimensional. In this paper, we propose an U-net and graph attention network based end-to-end unsupervised deep residual generative adversarial network (GAT-URGAN). We add residual and attention blocks for excavating non-local detail information. In addition, for threedimensional structure, we introduce graph attention layers to learn global structural detail information of inter-slices. To handle unpaired training data, we utilize deep network prior and introduce dual adversarial loss to constrain the generation of noise. We perform extensive experiments on Brainweb MRI datasets, which are divided into single noise images and mixed noise images, and accomplish reformative consequences compared to recent state-of-the-art denoising approaches. Keywords Unsupervised · Image denoising · Medical image · Graph attention network
1 Introduction Image noise is a momentous element that deteriorates image quality and reduces the accuracy of computer vision assignments [11, 13], like semantic segmentation [15] and object detection [18]. which is more distinct in medical images. Computer aided diagnosis (CAD) has been paid more attention to the development of deep learning. However, people attach more attention to the accuracy and reliability of diagnosis for the particularity of medical applications. Medical image noises are bound to affect further analysis such as the detection of lesion and tissue segmentation. There are a lot of image denoising methods including supervised and unsupervised methods. Supervised methods are hard to apply for that medical images are more complicated and are difficult
Xiang Pan
[email protected] Tianxu Lv [email protected] 1
The School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, 214122, China
2
The College of Life Information Science and Instrument Engineering, Hangzhou Dianzi University, Hangzhou, 310018, China
to be annotated. Meanwhile, the prior unsupervised methods are always aimed to deal with natural two-dimensional images. Therefore, prior proposed denoising methods are not well suitable for medical images, which are always three-dimensional. Traditional deep neural network (DNN), such as convolutional neural network (CNN) [12] is often used to deal with Euclidean Structure data, which cannot apply for NonEuclidean Structure. Graph Neural Network (GNN) has been proposed to handle irregular data, such as social network data and graph one. Based on graph data structure, GNN equates each neuron with a vertex and each connection between two neurons with an edge between two vertexes. Kipf & Welling [17] put forw
Data Loading...
