Recognizing COVID-19 positive: through CT images

Wu, X. Y.; Wang, Z. H.; Hu, S. C.; Ieee,

Since the beginning of 2020, the COVID-19 infection caused by a virus called SARS-CoV-2 has spread rapidly around the world. Recently, researchers and public health officials from different disciplines studied the pathogenesis of SARS CoV 2 and found that the imaging pattern of patients with SARS CoV 2 infection had been observed on computed tomography (CT). This article is to measure whether the traditional deep learning algorithm can rely solely on lung CT images as a basis for the presence of new coronary pneumonia. Using the classic deep learning algorithms of AlexNet, VGG, ResNet, SqueezeNet and DenseNet as the basis, using the lung CT data of patients with new coronary pneumonia published on Kaggle as training and testing, and testing whether the pretraining migration learning method will Make the algorithm get a higher accuracy rate. According to the results, the accuracy rate of all algorithms without the pre-training model is more than 70%, and the accuracy rate of some algorithms reaches 82%. It shows that the deep learning algorithm, driven by a small amount of data, can not be completely used as a means of identification, but the algorithm using deep learning can help doctors identify. Moreover, with the increase of data, a more optimized learning algorithm can also obtain higher accuracy.