Covid-19 Chest X-Ray Detection Using 3 Class Classification

Abstract

The Corona virus (COVID-19), a member of the virus family, first appeared in Wuhan, China, in December 2019. The COVID-19 disease can cause different symptoms, including high fever, cough, diarrhea, weakness, and respiratory problems. COVID-19 may lead to severe problems. In some active cases, Patients may experience difficulties breathing, pneumonia, sudden cardiac arrest, multi-organ failure, and death. We can use Deep Learning (DL) technology to detect and predict this disease using Chest X-Ray images and classify Covid-19 patients whether they have or may not have this fatal disease in the future. This study presents several Deep Neural Networks (DNN) for diagnosing early identification of Covid-19 to save doctors and radiologists money and time. Because Deep Learning (DL) algorithms perform well with large amounts of data, it is helpful for diagnosing the Covid-19. In this report, the dataset took from the Kaggle. Where the COVID class has 3616 images, Viral Pneumonia has 1345 images, and CN has 10192 images. This dataset is imbalanced, so I need to make a new dataset where I took 1655 pneumonia images from the Kaggle Pneumonia dataset and set all the class images to 3000. Here, training and validation datasets are divided into 80% and 20% proportions. The DNN models include the variation in DenseNet169, DenseNet201, and ResNet152, along with the KL divergence loss function. A comparison is made with these three base models. The loss function reduces the error and improves accuracy, giving an accuracy of 98.15% for training, and 95.72% for the validation dataset.