Abstract：Objective To investigate the value of CT images texture analysis in differential diagnosis of corona virus disease 2019 and other similar pneumonias. Methods The CT plain scan images of 33 patients with COVID-19 and 33 patients with other
similar pneumonias were mixed. The maximum level of lesion was selected and imported into Mazda 4.6 software and ROIs were manually drawn. The optimum texture parameters were selected through Mutual information (MI), Fisher coefficient , probability
of classification error and average correction coefficient (POE + ACC) and the combination of three methods (Fisher + POE+ ACC + MI) respectively. The model of artificial neural network was builded. The diagnostic results of CT texture analysis and radiologist
results were compared with the clinical results. And misdiagnosis rates were analyzed between texture analysis and radiologist. Results The tatal misdiagnosis rate of MI texture analysis was 34.85% (23/66), Fisher was 25.76% (17/66), POE+ ACC was
19.70% (13/66), the combined value of the three methods was 10.60% (7/66), and the rate of misdiagnosis by imaging physicians in identifying COVID-19 and other similar pneumonia was 31.82% (21/66). Misdiagnosis rate of Fisher + POE+ ACC + MI was lowest.
Meantime, the misdiagnosis rates of other pneumonias were significantly higher than that of COVID-19 read by radiologists (P<0.05).Conclusion CT images texture analysis can be used to differential COVID-19 and other similar pneumonias.
雷爱春，李光芒，王志兵，孙攀，储燕. 新型冠状病毒肺炎和其他相似肺炎CT平扫图像的纹理分析[J]. 中国医疗设备, 2020, 35(12): 91-94.
LEI Aichun, LI Guangmang, WANG Zhibing, SUN Pan, CHU Yan. Texture Analysis of CT Plain Scan Images of COVID-19 and Other Similar Pneumonias. China Medical Devices, 2020, 35(12): 91-94.
