COVID-19 and Non-COVID-19 Classification from Lung CT-Scan Images Using Deep Convolutional Neural Networks

Authors : Özlü Dolma

Pages : 53-60

View : 63 | Download : 68

Publication Date : 2023-12-19

Article Type : Research Paper

Abstract :In this study, three different convolutional neural network (CNN) architectures have been used for SARS-COV-2 infection (COVID-19) detection from lung Computerized Tomography (CT) scan images. The dataset comprises 2481 lung CT-scan images, of which 1252 are positive for COVID-19 infection. First, a simple CNN, LeNet-5, was trained from scratch, which resulted in poor classification performance with an accuracy value of 0.78. Then, to overcome the drawback of the limited availability of data, the convolutional bases of two pre-trained networks, VGG-16 and MobileNet, were leveraged to extract features from the dataset. On top of the feature extraction outputs, new classifiers were trained. When the VGG16 and the MobileNet CNN’s convolutional bases were used for feature extraction, accuracy values of 0.974 and 0.984 were obtained, respectively. The findings indicate that using pre-trained CNN models for feature extraction and then training a simpler, fully connected network structure for classification successfully differentiates CT-scan images of patients with COVID-19 infection from the ones without COVID-19 infection.
Keywords : COVID 19, Convolutional Neural Networks, Feature Extraction, Classification, Deep Learning