Enhanced Medical Data Classification Using Deep Convolutional Generative Adversarial Networks

Abstract

In today’s world, where there is an abundance of data all around us, one field sharply lacks the presence of data, and that is the medical field. Being of a sensitive and time critical nature, it is important to create methods that aid and speed up the identification of diseases and disorders using medical data. But for that, the pre-requisite is to have easy access to such data. There has been much work on supervised learning using CNNs, but the full potential of unsupervised learning with Generative Adversarial Networks (GANs) has not been explored, despite its ability to be a game changer in the field of computer vision. In this research paper, we propose a data augmentation and classification scheme using a two-fold architecture containing a Deep Convolutional Generative Adversarial Network (DCGAN) and a Convolutional Neural Network (CNN). We use a dataset of Chest X-Rays containing two classes, Normal and Pneumonia, to synthesize new images using the DCGAN. The generated images are then appended to the existing dataset and the new hybrid (synthetic + original) dataset is classified using the CNN. It is observed that the accuracy of the classifier significantly increases as compared to the accuracy pre-augmentation.