Deep Learning-Based Recognizing COVID-19 and other Common Infectious Diseases of the Lung by Chest CT Scan Images

Min Fu; Shuang-Lian Yi; Yuanfeng Zeng; Feng Ye; Yuxuan Li; Xuan Dong; Yan-Dan Ren; Linkai Luo; Jin-Shui Pan; Qi Zhang

The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10. 1101 /2020 Performance of the AI algorithm during test 250 From Zhongshan Hospital Xiamen University, and the fifth Hospital of Wuhan, 29201 251 CT scan images were collected from the following patients: 50 cases of COVID-19 252 pneumonia, 52 cases of non-COVID-19 viral pneumonia, 53 cases of bacterial 253 pneumonia, 54 cases of pulmonary tuberculosis, 100 cases of normal lung, were 254 employed to develop the model (Table 1) . Multi-class comparison was performed 255 between COVID-19, non-COVID-19 viral pneumonia, bacterial pneumonia, 256 pulmonary tuberculosis, and normal lung. Confusion matrix of the AI framework 257 based on test dataset was shown in Figure 2 . Binary classification between COVID-19 258 and the other four types, including non-COVID-19 viral pneumonia, bacterial 259 pneumonia, pulmonary tuberculosis and normal lung, was also implemented to 260 evaluate the performance of recognizing COVID-19. For test dataset, accuracy, 261 sensitivity, specificity, PPV, and NPV of recognizing COVID-19 were 98.8%, 98.2%, 262 98.9%, 94.5%, and 99.7%, respectively (Supplementary Table 2 ). The ROC curve was 263 generated to evaluate the AI algorithm's ability to distinguish COVID-19 from other 264 four types. The area under the ROC curve was 99.0% (Figure 2 the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.