DeepAttentiveNet: An automated deep based method for COVID-19 diagnosis based on chest x-rays

Abstract

The recent outbreak of coronavirus has impacted the whole world. The infectious respiratory disease has killed millions of people all over the world. The process of detecting the disease through RT-PCR and other tests is very time-consuming, and testing kits are not widely available. Chest x-rays and chest CT scans are also very effective techniques for diagnosing respiratory diseases. This paper proposes a DeepAttentiveNet, a deep-based architecture that applies the pre-trained CNN-based architecture DenseNet to extract the spatial features from the images. This is followed by the attention mechanism, which focuses on the information-rich region on the images, thus enhancing the overall classification process. The performance of our model is analyzed on the COVID 19 Radiography dataset, which contains 21,000 x-ray images corresponding to different respiratory infections like COVID 19, lung opacity, and viral pneumonia. Hence our model can categorize the x-rays with a 97.1% F1 score and 97.5% accuracy. We have also compared our architecture with other popular CNN-based models and baseline methods to demonstrate the superior performance of the model.