Synthetic data generation for medical by generative adversarial network
Yaoxuan Liu, Hao Xu, Ziling Tang · 2022 International Conference on Electronics and Devices, Computational Science (ICEDCS) · 2022
As the world develops, medical data privacy is becoming an important issue. Differential privacy can better ensure accuracy and reduce the reflection of personal privacy, while generative adversarial networks can generate data sets. Obviously, the combination of differential privacy and generative adversarial network (GAN) can synthesize sensitive medical data to form new data sets, which not only protects personal privacy but also does not affect medical research. We summarize the current difficulties in medical data, compare the common medical data types with typical DP methods of different GAN, analyze the advantages and disadvantages of each method, and discuss future development. At the same time, we also introduce GAN applied in the image data field. Because this approach is very mature, we only introduce some commonly used GANs, compare their improvements, and introduce their applications in data generation. This may help in the future development of medical data privacy algorithms.