Urine Red Blood Cells Generation Using StyleGAN2 Network

Abstract

With the development of computer hardware and deep learning theory, the application of deep learning in the field of digital medicine is very popular. Deep learning technologies use neural networks to build the relationship between medical data and disease diagnosis from large amounts of data. The acquisition and labeling of medical data has always been very time-consuming and laborious. In addition, the law states that patient data belongs to personal privacy, so medical data is rarely disclosed. Based on the urine red blood cell dataset (URBC), we train the recently proposed StyleGAN2 network to generate a public dataset S2RBC-256. It effectively avoids the strict regulations on patient ethical privacy and hospital data. This dataset can further help future scientific research in the related fields of red blood cell. In terms of digital metrics (SSIM, PSNR), the images we generate achieve good results. Through the review of medical experts, our generated images are visually very consistent with real images. Therefore, the public dataset we generate can be used for training of deep neural networks for the purpose of reducing the labor and material consumption of annotation.