COVID-19 and Pneumonia Detection using Hybrid VGG-16 model using Chest X-rays

ABSTRACT

The time frame of 2020 to present day 2022 primarily highlights the COVID-19 pandemic. The humanity is being largely affected by SARS-CoV-2(The Severe Acute Respiratory Syndrome CoronaVirus 2) because of its highly infectious characteristic which can be even fatal in severe cases. The World Health Organization (WHO), have reported over 544.3 million verified cases of COVID-19 globally till date, including over 6.3 million deaths. The reason why SARS-CoV-2 is considered to be a dangerous illness is due to this relatively high mortality and contagious rates, in addition to asymptomatic individuals also being carriers of the virus. The only way to identify susceptible populations and to attempt to control the spread would be via RT-PCR COVID testing of all individuals, which is time consuming and expensive. The challenges of this testing mechanism and the prolonging end of the pandemic are the primary motivation to bring up an effective system over a large test cases with a reduced time constraints. This paper proposes a combination of the pretrained convolutional neural network, VGG-16(Visual Geometry Group-16) and GRU(Gated Recurrent Unit) to differentiate the Pneumonia and COVID-19 attack from chest X-rays(CXRs). The proposed model employs VGG-16 to extract features from the CXR inputs, and the GRU classifies it. We experimented this model over 6939 CXR images with 3 classes (COVID-19, Pneumonia, and Normal) and the training produced encouraging macro average precision, recall, and f1-score of 0.9525, 0.9524, and 0.9524 respectively. These results indicate hybrid deep learning systems can greatly aid in the early detection of COVID-19 using CXRs and thereby reduce the widespread of the pandemic. We believe that early diagnosis can be easily and effectively done using this model. © 2022 IEEE.