SmartCovNet: An Intelligent ResNet-50 and Novel Neural Classifier Structure for COVID-19 Discovery Via CT Imageries

The COVID-19 pandemic has necessitated immediate attention towards the development of novel structures. The RT-PCR test is the widely used technique for diagnosis; however, it poses disadvantages such as longer delays and inconsistency in test results. We present a SmartCovNet framework to distinguish COVID and non-COVID cases using CT scan images. An integration of a pretrained ResNet-50 architecture with a novel neural network classifier is employed. The lung regions are highlighted by employing adaptive histogram equalization with contrast restriction. After extracting the features from ResNet-50, we pass them to the novel neural network classifier. SmartCovNet has achieved a classification accuracy of 99.53%, along with high specificity, sensitivity, and F1 score. The performance metrics indicate the effectiveness of SmartCovNet in the diagnosis of COVID infection.