Utilizing Convoluted Neural Networks to predict COVID Infections in Individuals

Abstract

The world is currently witnessing a second wave of the novel Coronavirus epidemic and countries, especially in the Indian Subcontinent have been heavily affected in its wake. With 158.4 million cases worldwide at this moment, it has not only put a tremendous pressure on the entire diagnostic process of COVID infections (using PCR, Rapid or Serology tests) in suspected individuals but has also put at risk it’s availability and affordability at a community level in certain countries.Hence, suspected individuals with mild symptoms are often requested to self quarantine and not take the COVID-Diagnostic tests so that the availability of tests can be ensured for more serious patients. There is also a certain level of hesitancy in patients to go for such diagnostic tests due to factors such as overcrowding, long queues and social stigma. Since one of the major symptoms of COVID infection is coughing, a system for recognizing and diagnosing COVID positive individuals based on raw cough data would have a multitude of beneficial applications at a personal and social level. In this work, we present a system that utilizes Convolutional Neural Networks (CNNs) and audio signal spectrograms  to diagnose any individual for Coronavirus infection based on their unique cough audio features. Our diagnosis model achieves an accuracy of about ___ . This result clearly shows that our single diagnosis model is capable of predicting COVID infections in suspected individuals which can serve as a pre diagnostic and preliminary assurance tool.