Efficient Classification Approach Based on COVID-19 CT Images Analysis with Deep Features
Mostafa A. Kamel, Mohamed Abdelshafy, Mustafa AbdulRazek, Osama Abouelkhir, Amr Fawzy, Ahmed T. Sahlol
Abstract
Currently, a new coronavirus(COVID-19) has affected millions of people worldwide. For this reason, it's not sufficient that radiologists can slow down the virus spreading manually. Convolutional Neural Networks (CNNs) can be utilized as a tool to aid radiologists in diagnosing COVID-19 images, which consequently can save efforts and time. In this work, a dataset of CT images of confirmed and negative COVID-19 was used for the screening of COVID-19. Some preprocessing operations were applied to enhance the COVID-19 CT images which aim at including only the Area of Interest (AOI). This was accomplished in three stages. First, a conversion of the CT images to the binary scale was performed by applying a global threshold algorithm. Then, the median filter algorithm was applied to remove random noise. Then, we include only the ROI (the lung) and exclude other parts of the images. Finally, we applied VGGNet 19 to extract features from the preprocessed CT images, which is a popular CNN architecture, trained previously on ImageNet. The proposed pipeline showed high performance by achieving 98.31%, 100%, 98.19% and 98.64% of accuracy, recall, precision and f1-score, respectively. To the best of our knowledge, these results are the best published on this dataset when compared to a set of recently published works. Also, the proposed model overcomes several popular CNNs architectures.