Towards Covid-19 Detection in X-ray images using Convolutional Neural Network

Over the past 18 months, COVID-19 is continuously causing deadly consequences and devastating economic situations around the world. The vaccinations are still not adequately mature and vaccinating all the people around the world is a long process. Moreover, there is a rapid generation of variants of COVID-19 that are difficult to handle promptly. The skyrocketing pandemic of COVID-19 has knocked down the regular lifestyle of everyone, from individuals to massive corporations. The health industry is overburdened by the increasing number of patients that they are receiving. Efficient and stable screening methods of this disease will help in speeding up the detection process and take appropriate measures to flatten the curve on this disease. There is a high demand for COVID-19 discovery frameworks where the existing approaches still have some drawbacks. The most commonly used methods such as RTPCR have drawbacks of high false-positive rates, costly special kits, providing delayed and less accurate test results. Therefore, other diagnostic methods are actively explored. In this scenario, Artificial Intelligence comes to utilization with its branch of deep learning. Convolutional neural networks (CNNs) are a class of deep learning that are utilized in analyzing images, including medical imaging like Chest X-rays. This paper aims to analyze the power brought by CNNs to medical imaging in solving the global problem of early identification of COVID-19 in potential patients by analyzing their X-ray scans. Moreover, the entire process of image acquisition and analysis, its positive impact by minimizing asymptomatic patient contact with others and increasing the efficiency of the work environment at the frontline hospitals is discussed. The test results in this study showed that an accuracy of 95.11% was obtained on a dataset of 2905 images and an accuracy of 96.07% was obtained on an augmented dataset of 9337 images to detect Covid-19 in X-ray images, thus proving that deep learning models work better if the size of the training dataset is increased.