CT Slice Segmentation for Patients with COVID-19 based on Transformer and Convolution Neural Network

ABSTRACT

In this paper, we propose a new CT slice segmentation method for patients with COVID-19 based on Transformer and Convolutional Neural Network (CNN), which will speed up the diagnosis of CT slices and improve the diagnostic efficiency. We built a network model called 'TRUNet'. This model combines the advantages of Transformer and CNN, not only can fully utilize the global context extraction of Transformer, but also combine the high resolution features learned by CNN. It improves the segmentation effect of lung CT slices to a certain extent. Experimental result shows that the method proposed in this paper can segment lung CT slices into four classes. On the test set, the mean IoU of the model reaches 0.8467. © 2022 IEEE.