Interpretable COVID-19 classification leveraging ensemble neural network and explainable AI

Abstract

COVID-19 which is also none as Corona Virus Disease is rst discovered in a city of China named Wuhan at December 2019 and it has been announced as a global pandemic at the middle of 2020. SARS-CoV-2 virus COVID-19 and that can also act as a trigger to cause respiratory tract infection ranging from mild to deadly. According to experts, this virus may also infect the upper respiratory system, which includes the sinuses, nose, and throat, as well as the lower respiratory system, which includes the windpipe and lungs. The disease can infect other people via respiratory droplets and coming near to the COVID-19 infected people as well as touching those objects of surfaces which are the virus contaminated. Nowadays, millions and millions of people across the globe are su ering from this disease causing a huge death rate. Even after taking serious precaution measures, the number of patients dealing with this disease and the death toll is still rising at a drastic rate. In this paper, we approach a fast and e ective measure to detect COVID-19 using CT scan images. First, we collected data and classi ed using VGG16, VGG19, E cientNetB0, ResNet50 and ResNet152. Form our result; we got an accuracy rate of 85.33% from VGG16, 87.86% from VGG19 and 82.35% from ResNet101. Then we formed an ensemble model with these best three classi ers and achieved a best overall accuracy rate of 90.89% from COV19EXAI V1 and 91.82% from COV19EXAI V2. Finally, we integrated XAI in our model to achieve a better understand of our classi cation.