Deep Learning Based on High-Dimensional Tensor for COVID-19 Diagnosis

The breakout and rapid spread of coronavirus disease 2019 (COVID-19) has become a global health concern. The disease has infected more than 18.7 million people and caused more than 707,000 deaths all over the world, as of August 6 in 2020. Computed tomography (CT) is promising to provide screening and testing for COVTD-19. This paper proposes cross-layer connection neural network based on high-dimensional tensor to classify the CT scans. Cross-layer connected network DenseNet and ResNet have sophisticated network structure residual block and dense block which perform well in extracting deep features. Specifically, we choose 1, 4, 8, 16 and 32 as input tensor dimensions to respectively conduct experiments on the two neural networks. Extensive experimental results show that when DenseNet-121 is the backbone network and the tensor dimension is 16, the experimental results are the best, and the accuracy and AUC are both 0.92, precision is 0.90, Fl is 0.95 and recall is 0.99.