Improved deep learning model for differentiating novel coronavirus pneumonia and influenza pneumonia

Min Zhou; Yong Chen; Dexiang Wang; Yanping Xu; Weiwu Yao; Jingwen Huang; Xiaoyan Jin; Zilai Pan; Jingwen Tan; Lan Wang; Yihan Xia; Longkuan Zou; Xin Xu; Jingqi Wei; Mingxin Guan; Jianxing Feng; Huan Zhang; Jieming Qu

The copyright holder for this preprint . https://doi.org/10.1101/2020.03.24.20043117 doi: medRxiv preprint instead, were misdiagnosed by three specialists. We have verified the clinical applicability of our developed AI model by including data from multiple machines and centers. We first demonstrated that the AI model performed well using training and test data from four machines of three centers with an AUC of 0.99. Similar performance of the model with specialists on independent verification data from fifteen machines of eight centers further suggests good clinical applicability. Although our AI system achieved good performance, it misclassified a small number of NCP and IP patients, which may be caused by poor spatial resolution of some of images. In this study, we used 5 mm instead of 1 mm layer thickness in CT reconstruction, which would limit our capability to detect small lesions. Nevertheless, 5 mm layer thickness is a standard parameter in most hospitals and is sufficient to identify major imaging differences between NCP and IP as demonstrated by our study. Therefore, it is worthy to sacrifice certain accuracy to provide wider applicability of the deep learning model.