Deep Learning Based Data Augmentation and Classification for Limited Medical Data Learning

Deep learning methods, especially convolutional neural network (CNN), which have made breakthroughs in many fields of computer vision, and one of the most prominent features of deep learning is that large-scale dataset with annotations should be used. However, obtaining such a dataset in the medical field is really a challenge due to scarcity of annotated data. In our work, a deep learning based framework is proposed to generate more real data from the existing data with Generative Adversarial Networks (GAN) and classify the suspicious lesions with CNN. The proposed method is applied on the diagnosis of diabetic retinopathy. The experimental results show that our deep learning framework achieve a better classification performance than the traditional deep learning methods.