Simultaneous Detection of Covid-19 and Its Pneumonia using Multitask Learning

Abstract

With the rapid growth of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) or Covid-19 into a pandemic, quick and efficient alternative testing methods were needed. Although Viral Nucleic Acid tests are the primary and standard method of testing, the time-consuming process, and the lack of availability of test kits in certain areas have been problematic for the quick diagnosis of the disease. Therefore, using radiologic modalities such as chest X- rays and Computerized Tomography (CT) were studied due to their wider availability because of their usage in the diagnosis of other diseases. This research is based on chest X-rays and tests the usage of deep multitask convolutional neural networks (CNN) to detect both Covid-19 and Covid-19 related pneumonia conditions in a patient simultaneously. Usage of chest X-rays allows for wider availability for usage in rural areas, where computerized tomography facilities are rare. Current results from separate custom CNN models with same layer structure but different task specific features, give an accuracy of 94% on Covid-19 detection and 90% accuracy on Covid-19 pneumonia detection. As a novelty, this paper suggests that a multitask learning based CNN model in the same architecture would be viable to detect both conditions from a single neural network, simultaneously. The simultaneous detection of Covid-19 and Covid-19 pneumonia in a patient is a further extension to traditional testing methods and allows for more effective treatments from the beginning.