Generalization of Deep Neural Networks for Chest Pathology Classification in X-Rays Using Generative Adversarial Networks

  • 2018-02-12 18:25:28
  • Hojjat Salehinejad, Shahrokh Valaee, Tim Dowdell, Errol Colak, Joseph Barfett
  • 1


Medical datasets are often highly imbalanced with over-representation ofcommon medical problems and a paucity of data from rare conditions. We proposesimulation of pathology in images to overcome the above limitations. Usingchest X-rays as a model medical image, we implement a generative adversarialnetwork (GAN) to create artificial images based upon a modest sized labeleddataset. We employ a combination of real and artificial images to train a deepconvolutional neural network (DCNN) to detect pathology across five classes ofchest X-rays. Furthermore, we demonstrate that augmenting the originalimbalanced dataset with GAN generated images improves performance of chestpathology classification using the proposed DCNN in comparison to the same DCNNtrained with the original dataset alone. This improved performance is largelyattributed to balancing of the dataset using GAN generated images, where imageclasses that are lacking in example images are preferentially augmented.


Introduction (beta)



Conclusion (beta)