Deep learning models for classifying cancer and COVID-19 lung diseases

Abstract

The use of Computed Tomography (CT) images for detecting lung diseases is both hard and time-consuming for humans. In the past few years, Artificial Intelligence (AI), especially, deep learning models have provided impressive results vs the classical methods in a lot of different fields. Nowadays, a lot of researchers are trying to develop different deep learning mechanisms to increase and improve the performance of different systems in lung disease screening with CT images. In this work, different deep learning-based models such as DarkNet-53 (the backbone of YOLO-v3), ResNet50, and VGG19 were applied to classify CT images of patients having Corona Virus disease (COVID-19) or lung cancer. Each model's performance is presented, analyzed, and compared. The dataset used in the study came from two different sources, the large-scale CT dataset for lung cancer diagnoses (Lung-PET -CT-Dx) for lung cancer CT images while International COVID-19 Open Radiology Dataset (RICORD) for COVID-19 CT images. As a result, DarkNet-53 overperformed other models by achieving 100% accuracy. While the accuracies for ResNet and VGG19 were 80% and 77% respectively.