Title: MS-PriGAN: prior knowledge-based multi-scale gradient generation adversarial network for ovarian cancer CT image generation
Authors: Xun Wang; Zhiyong Yu; Lisheng Wang; Mao Ding
Addresses: College of Computer Science and Technology, China University of Petroleum (East China), Qingdao, 266580, Shandong, China ' College of Computer Science and Technology, China University of Petroleum (East China), Qingdao, 266580, Shandong, China ' College of Computer Science and Technology, China University of Petroleum (East China), Qingdao, 266580, Shandong, China ' Department of Intensive Care Unit, The Second Hospital of Shandong University, Jinan 250033, China
Abstract: Despite the great success of deep neural networks in medical lesion detection tasks, their adaptation to CT images of ovarian cancer is known to be difficult, and the most important reason is the lack of ovarian cancer datasets. A commonly accepted reason for this deficiency is that few people are willing to disclose CT imaging data due to privacy concerns, despite the fact that ovarian cancer is the first most prevalent cancer among women. In this work, we propose a method for generating synthetic high-resolution medical images using generative adversarial networks. The method addresses the instability of previous work in generating work on small-scale datasets by a simple and effective technique to acquire prior knowledge and achieve multi-scale gradient flow. We show that the proposed method still achieves good results even when synthesised on small-scale data, and with FID scores better than the current top-performing networks (GANs).
Keywords: deep learning; priori knowledge GAN; CT images; ovarian cancer.
International Journal of Adaptive and Innovative Systems, 2022 Vol.3 No.2, pp.135 - 143
Received: 11 Sep 2021
Accepted: 10 Dec 2021
Published online: 25 Jul 2022 *