transfer learning-based detection of covid-19 cases from chest ct scans

When taking into account the prevailing covid-19 pandemic scenario, by early detecting covid-19, we can formulate an effective treatment plan and make decisions regarding disease containment. as a result of this issue, artificial intelligence (ai) specialists have been encouraged to develop models that employ deep learning techniques in covid-19 detection. these models diagnose infection severity rapidly and economically. the present study proposes a deep convolutional neural network (cnn) model based on pso, which helps identify covid-19 infections by chest computerized tomography (ct) scans. moreover, we demonstrate how pre-trained models can classify the disease through transfer learning. initially, the random search is used to identify an optimal cnn model. the transfer learning strategy presents an analysis of several popular pre-trained models. the optimal cnn model inherits several layers from these previously trained models, and we then fine-tune the selected optimal cnn model accordingly. the proposed architecture is built using three pre-trained models with the highest quality. pso algorithm is applied to estimate how each pre-trained model will affect the ultimate detection of the suggested model. to train the model, we analyzed two publicly available datasets—covid-ct and sars-cov-2—applying distinct pre-processing techniques to each. according to the experimental results, our pso-based configuration optimization performed well on this dataset and can achieve better results with more training data. as a result of extensive parameter tuning, the proposed model can identify covid-19 with an accuracy of up to 90.32%. this model will facilitate the detection and diagnosis of covid-19 promptly.