Label generation system based on generative adversarial network for medical image

Abstract

In recent years, the generation model has made great progress in the task of less label sample data. Aiming at the heavy task, high cost, time-consuming and laborious problems of medical image labeling, this paper proposes an image label generation model based on generative adversarial network (GAN). The generator consists of a convolution network and a long-term and short-term memory network. It generates a text description for the input image. At the same time, the discriminator consists of a convolution network, calculates the difference between the generated description and the real description, and transfers the gradient to complete the confrontation training. In this paper, the model is trained on the INbreast dataset, and the experiment show that the model achieves good results in the generation of medical image data labels.