CLASSIFICATION OF CHEST X-RAY IMAGES USING DEEP CONVOLUTIONAL NEURAL NETWORKS

Abstract

In this paper a comparison between three different types of trained VGG convolutional neural networks (CNNs) is proposed for the classification of a pediatric chest X-ray image data set. A deep convolutional neural network with an architecture resembling the VGGNet is presented using dropout, decay and data scaling. Since the dataset had a class imbalance, this was solved using a simple method called data scaling. The training of the neural network was done using small batches with a binary cross entropy loss function. The same neural network was then implemented adding batch normalization layers, and comparisons were made. Furthermore, the chest X-ray dataset was also trained using transfer learning with a pre-trained neural network VGG16 on the ImageNet dataset. Later on juxtapositions were made on using both techniques. Additionally, in applying these methods we were able to achieve a classification with the accuracy higher than 0.95 and 0.97 for the training and validation datasets, whilst incorporating only 30 epochs.