Data Augmentation Using Auxiliary Classifier Generative Adversarial Networks

Computer-aided diagnosis technology based on convolutional neural networks (CNN) is one of the focuses of medical imaging research. However, it is not easy to obtain medical image data, which brings great difficulties to computer-aided diagnosis. To address this issue, we adopt a data augmentation method that uses auxiliary classifier GANs (ACGAN) to synthesize medical images. ACGAN conducts data training on the existing medical image dataset. The generator and discriminator learn the representation level of the image from the local features to the overall scene. Output the new medical image data combined with the features of the dataset, and output the classification of the image. By using transposed convolution in discriminator network to replace upsampling, improve its ability to extract features; Preprocess the image, improve the image quality, enrich the amount of information, and strengthen the image recognition effect, so as to improve the ability of the generator network to generate data and the accuracy of the discriminator network classification. Experimental results show that our proposed ACGAN framework can achieve effective learning of image features, and the quality of the generated data is high. It can effectively balance medical image data and further improve the accuracy of auxiliary diagnosis. We hope that this method can make up for the lack of image data, improve the accuracy of image recognition, and accelerate the development of computer-aided diagnosis technology.