COVID-19 detection from lung CT-scan images using transfer learning approach

Abstract Since the onset of 2020, spread coronavirus disease (COVID-19) has rapidly accelerated worldwide into a state severe pandemic. COVID-19 infected more than 29 million people and caused 900 thousand deaths at time writing. it is highly contagious, causes explosive community transmission. Thus, health care delivery been disrupted compromised by lack testing kits. COVID-19-infected patients show acute respiratory syndrome. Meanwhile, scientific involved in implementation deep learning (DL) techniques to diagnose using computed tomography (CT) lung scans, since CT pertinent screening tool due its higher sensitivity recognizing early pneumonic changes. However, large datasets CT-scan images are not publicly available privacy concerns obtaining very accurate models become difficult. overcome this drawback, transfer-learning pre-trained used proposed methodology classify (positive) (negative) patients. We describe development DL framework that includes (DenseNet201, VGG16, ResNet50V2, MobileNet) as backbone, known KarNet. To extensively test analyze framework, each model was trained on original (i.e. unaugmented) manipulated augmented) datasets. Among four KarNet, one DenseNet201 demonstrated excellent diagnostic ability, with AUC scores 1.00 0.99 for unaugmented augmented data sets, respectively. Even after considerable distortion dataset) achieved an accuracy 97% dataset, followed MobileNet, VGG16 (which accuracies 96%, 95%, 94%, respectively).