Deep learning-based COVID-19 detection: State-of-the-art in research

In the last two years, the coronavirus (COVID-19) pandemic put healthcare systems around the world under tremendous pressure. Imaging techniques (like Chest X-rays) play an essential role in diagnosing many diseases (such as COVID-19). There have been intelligent systems (Machine Learning (ML) and Deep Learning (DL)) able to identify COVID-19 from similar normal diseases. In this paper, we start by overviewing the status of COVID-19 from a historical standpoint and diagnosis updates. Moving on, provide an overview of the convolutional neural networks. Then, we elaborate Transfer learning method and its main approaches. Next, we provide a critical literature review on implementing Deep learning techniques: 1) Novel deep learning architecture; 2) Direct use of deep learning; 3) Transfer learning fine-tuning technique, and 4) Transfer learning feature extraction technique. For each of these, we evaluate and compare very recent studies published in highly ranked journals. The experiments show that all techniques achieve closer accuracy, ranging from (98-100 \%). Along with all, the direct use of the deep learning technique records the highest accuracy and is less time-consuming and resource spending. Therefore, establishing such a technique is useful to predict the outbreak early, which in turn can aid in controlling the pandemic effectively.