Classification of COVID-CT Images Utilizing Four Types of Deep Convolutional Neural Networks

Abstract

In this paper a method is presented for the classification of COVID-CT (CT_COVID, CT_NonCOVID) image data set. Four different types of deep convolutional neural networks are proposed, two with the architecture resembling the VGGNet, one resembling the LeNet-5 and one using transfer learning. In addition, neural networks utilized the following techniques: decay, dropout and batch normalization. Since we needed to combat a significantly small dataset, we used data augmentation in order to transform and expand our dataset. Moreover, juxtapositions were made when observing the results given by these four neural networks, as well as the affect made by two different optimizers. The training of the neural networks was done using small batches with a binary cross entropy loss function, in order to achieve an up to scratch classification accuracy.