A novel internet of things based on deep neural network framework using soft-attention convolutional neural networks for COVID-19 detection

Background: The impact of coronavirus (COVID-19) pandemic on health care is universal. The risks resulting from emerging contagious viruses and the efficacy of vaccines are persisting due to the presence of different variants. Learning of deeper and more interpretable models from COVID-19 data are conducive to understand this disease and to study the virus spread, individual diagnosis and may be other engrossing relating issues. However, some difficulties and intricacies are arising from the scarcity of precisely labelled data. Previous works have exploited existing Deep Neural Network (DNN) models that are pre-trained on large datasets like ImageNet. Method: In this paper, a new framework is proposed in order to monitor and predict COVID-19 cases and other diseases, pursuing medical data. The currently proposed framework essentially relies on (1) an Internet of Things (IoT) processing model to collect data and operate on them later, (2) a DNN model for data processing, known as REGATT. This proposed model is based on a pre-trained REGNet model finely tuned by spatial, channel ATTention and convolutional layers, boosting feature representation and discrimination. Results: Comparative experimental results on four different benchmark datasets show that the proposed model leads to a promising solution for diagnosing COVID-19. Conclusion: It is concluded that an IoT and DNN-based solution are a viable way for the diagnosis of not only COVID-19 but also other diseases. It is advisable that future works explore the development of interpretable models.