Fast automated detection of COVID-19 from medical images using convolutional neural networks
Abstract Coronavirus Disease 2019 (COVID-19) is a global pandemic that poses significant health risks. The sensitivity of diagnostic tests for COVID-19 is low due to irregularities in the handling of the specimens. We propose a deep learning framework that identifies COVID-19 from medical images as an effective auxiliary testing method to improve diagnostic sensitivity. We use pseudo-coloring methods and a platform for annotating X-ray and computed tomography (CT) images to train and evaluate the convolutional neural network (CNN). The CNN achieves a performance similar to that of experts and provides high scores for multiple statistical indices, with F1 scores above 96% and specificity over 99%. Heatmaps are used to visualize the salient features extracted by the CNN. The CNN-based regression provides strong correlations between the lesion areas in the images and five clinical indicators, improving the interpretation accuracy of the classification framework. The proposed method represents a potential computer-aided diagnosis method for COVID-19 in clinical practice.