The Clinical Subtypes of Corona Virus Disease 2019 Correspond to CT Findings and the Value of Artificial Intelligence

Objective:To investigate the novel coronavirus pneumonia clinical manifestation of CT and analyze the application value of AI. Methods:A retrospective analysis of 44 cases of the novel coronavirus pneumonia with complete clinical and CT data from January 23, 2020 to February 25, 2020 with the help of "uAI novel Coronavirus Pneumonia Intelligent Assisted Analysis System". Results:1.4 cases were clinically mild, and chest CT showed negative findings. In 32 cases of clinical common type, CT showed multiple lesions of one or two lungs, which were flake like or wedge-shaped ground glass shadow, in which blood vessels and bronchi passing through. Most of them accompanied by interlobular septal thickening, paving stone sign and bronchi inflation sign, and in some lesions, there were small round lung consolidation. There were 8 cases of clinical severe and critical severe cases. CT showed a wide range of lesions, ground glass shadow, consolidation and fiber cord mixed, most of them were accompanied with "paving stone" sign and broncho inflation sign. 2. Compared with the visual inspection, the scope of the lesions in the artificial intelligence aided diagnosis software has a better consistency. The total lesion volume was shown to be 0 in 4 mild patients. The total lesion volume of 32 patients with clinical common type was(109.9 ±94.9) cm³. The volume of ground glass shadow in the lesion was(55.3 ±50.4) cm³,and the lung consolidation volume was(34.9 ±35.0) cm³. The total lesion volume of 8 patients with clinical Severe and critically type was(858.1 ±351.9) cm³. The volume of ground glass shadow in the lesion was(486.7 ±204.0) cm³. and the lung consolidation volume was(204.1 ±119.3) cm³. There was significant difference in the total volume of lesions, the volume of internal ground glass and the volume of consolidation between the general and severe/critical patients. The proportion of lesions in each lung lobe provided by artificial intelligence aided diagnosis software was statistically analyzed. The score of lung injury in clinical common type patients was(3.8 ±1.2) points, and that in severe/critical type patients was(10.4 ±5.1) points. The difference between the two was statistically significant(P<0.05). ROC curve showed that when the lung injury threshold was 5.5, the prediction of clinical classification was the highest, and the area under ROC curve was 0.996. Conclusion:CT scan of novel coronavirus pneumonia showed multiple ground glass opacities in the periphery of lung or under pleura, which contained normal vessels and bronchus. The enlarged, increased or consolidated lesions represented the progress of the disease.AI-assisted technology can effectively identify novel coronavirus pneumonia lesions, provide data related to total lesion volume, internal ground glass shadow and real volume, accurately quantify the degree of lung injury, provide help for clinical condition assessment, and improve the work efficiency of imaging physicians.