Deep LIFT incorporated Deep Learning Framework for Classification of Novel Coronavirus (COVID-19) using Computed Tomography scan Images

Abstract

The world was devastated by the Coronavirus disease (COVID-19) since 2019 and its variant pandemic. It is very important and crucial to detect coronavirus infected patients as early as possible. Researchers have proposed many neural based learning based methodologies based on Computed Axial Tomography scan (CAT scan) figures and X-ray figures to assist healthcare expert. In this paper, we have proposed classification of Covid-19 disease from CAT  study figures using hybridization of deep shift invariant based pretrained models and features generated by DeepLIFT (Neural based learning Important Features) technique. DeepLIFT assigns contribution scores based on the difference between each neuron's activation and its "reference activation." At first, picture preprocessing is done by resizing, normalization and then, features are extracted by hybridization of pre-trained deep models and DeepLIFT technique. Finally, images are classified into two categories, normal and Covid-19. Experiments are carried out on two standard COVID-19 CAT scan datasets and results show that the proposed hybridization technique gives better result than conventional one. Average accuracy of 95.4%, precision of 98.93%, recall of 95.96%, specificity of 80.12% and F1 score of 97.42 is achieved.