GAN-based one dimensional medical data augmentation

Abstract With the continuous development of human life and society, medical field is constantly improving. However, modern medicine still faces many limitations, including challenging previously unsolvable problems. In these cases, artificial intelligence (AI) can provide solutions. The research application generative adversarial networks (GAN) are a clear example. While most researchers focus on image augmentation, there few one-dimensional data augmentation examples. radiomics feature extracted from RT CT images data. As far as we know, first to apply WGAN-GP algorithm generate in field. this paper, input portion original real samples into model. model learns distribution generates synthetic with similar data, which solve problem obtaining annotated samples. We have conducted experiments public dataset Heart Disease Cleveland private dataset. Compared traditional method Synthetic Minority Oversampling Technique (SMOTE) common GAN for our has significantly improved AUC SEN values under different proportions. At same time, also shown varying levels improvement ACC SPE values. This demonstrates that effective feasible.