Covid CT-net: A deep learning framework for COVID-19 prognosis using CT images

Abstract

In 21st century, the surge of novel coronavirus (COVID-19) with its origin in Wuhan city of south China has caused a devasting effect not only on the public health but also on the economy of the countries all over the world. Early identification of the disease is the only significant way of combatting with COVID-19 infection. Though RT-PCR (Reverse Transcription Polymerase Chain Reaction) is the basic approach adopted, it has certain limitations such as less sensitivity, consumes more time and availability of limited number of kits. Therefore, analysis of radiological images using deep learning approaches is used as an alternative way to recognize the coronavirus epidemic at the initial stage. Most of the existing works use pretrained Convolutional Neural Network (CNN) prototypes and extremely large processing resources for the prognosis of covid infection from medical radiography images of lung. In this research, a novel framework utilizing deep convolutional neural network consisting of three convolutional layers, three max-pooling layers and one fully connected layer and RMSprop optimizer has been proposed to interpret COVID-19 from CT (computer tomography) images. Further, the efficiency of the proposed work has been enhanced by using the techniques such as data pre-processing and data augmentation. The efficacy of the proposed deep CNN model along with pre-trained CNN models such as DenseNet121, VGG16 (Visual Geometry Group), MobileNetV2, Xception and InceptionV3 has been evaluated on 1252 COVID CT images and 1240 non-COVID CT images. Furthermore, the empirical outcomes show that the suggested deep CNN is a robust approach and achieves better accuracy than the other competitive methods.