FiltDeepNet: Architecture for COVID Detection based on Chest X-Ray Images

Predicting the further evolution of a natural phenomenon is of great challenge, particularly with an epidemic. This is evident in the wide spread of COVID-19 caused by coronavirus due to which the world is agonizing since January 2020. The seriousness of the infection caused by coronavirus is not known completely, but the immune system is compromised in people with chronic respiratory or systemic diseases. But the good part is that majority of the infections are very mild and self-treated, and prediction of virus spread might be misleading due to different tests performed only on symptomatic patients. This motivates to explore various mathematical models to analyze the future course of the pandemic so that effective control strategies are put in place. In this paper, a deep learning architecture based on image filtering is proposed to control the ill effects of the pandemic caused by coronavirus. The proposed algorithm leverages image processing and deep learning algorithms on chest X-ray images and differentiates between the four classes: normal, covid affected, lung opacity, and viral pneumonia. A combination of five image processing filters with five different deep learning architectures is considered, and the best combination with better classification accuracy of COVID-19-affected chest X-ray images is chosen for COVID detection. The accuracy obtained is 96.1% on validation data and proves that combining a suitable image processing algorithm with a deep learning architecture would aid in improving the accuracy of image classification, particularly COVID detection.