Automatic COVID-19 Detection using Deep Features

Abstract

The disease named Coronavirus (COVID-19), caused by Severe Acute Respiratory Syndrome Coronavirus-2 (SARS CoV-2), first appeared in Wuhan city of China in late 2019, spread globally, infected millions with high mortality rate. With high transfer rate from one person to many others, this epidemic imposed an astounding impact on general wellbeing, everyday lives, and the worldwide economy. Traditional way of diagnosing COVID-19 is the testing toolkits is a time taking process. There is an urgent timely need for medical health care systems and precise identification of COVID-19 positive cases not only to treat the infected patients but also, to control its further spread. However, study of images obtained using radiology imaging techniques shows that the diseases such as COVID-19 can infect the respiratory organs such as lungs. These images contain reliable information to diagnose such diseases. This turned the professionals to look towards the use of advanced AI (Artificial Intelligence) techniques like DL (Deep Learning) coupled with CT (Computed Tomography) scans and x-rays to detect this disease, timely and efficiently. Fast and reliable feature learning ability of DL enables us to diagnose COVID from CT images automatically. However, to train a Deep Learning model, we need very large datasets to enable these models so that they will correctly perform required prediction on unseen datasets. To remove this dependency on data, scientist developed another technique called Transfer Learning which uses pre-trained state-of-the-art deep learning models for new type of datasets. Purpose of this research work is to propose a Transfer Learning based Deep (CNN) model to detect and differentiate between COVID-19 infected patients, normal, and other bacterial pneumonia patients in a timely and efficient way. Successful implementation of proposed model will reduce diagnostic time that is very important for saving lives.