Author(s)

Xin Yuan

Corresponding Author

Xin Yuan

Abstract

With the global spread of the COVID-19 epidemic, the diagnosis method of COVID-19 based on CT images has received extensive attention. This paper studies the diagnostic imaging method for COVID-19 based on 3D U-Net. The main contents include the introduction of the basic theory and structure of 3D U-Net, the analysis of the characteristics and modeling methods of COVID-19 infections, the optimization strategy of the convolutional neural network, the comparison of different loss functions and principal component analysis on the segmentation results, and the evaluation of the influence of noise error on the segmentation performance. This paper concludes that the COVID-19 imaging diagnosis method based on 3D U-Net can effectively identify and segment the focus of COVID-19, improving the accuracy and efficiency of diagnosis. It is of great importance for preventing and controlling epidemics and public health protection.

Keywords

COVID-19; CT images; 3D U-Net; Diagnostic imaging