Artificial intelligence for rapid identification of the coronavirus disease 2019 (COVID-19)

Xueyan Mei; Hao-Chih Lee; Kaiyue Diao; Mingqian Huang; Bin Lin; Chenyu Liu; Zongyu Xie; Yixuan Ma; Philip M. Robson; Michael Chung; Adam Bernheim; Venkatesh Mani; Claudia Calcagno; Kunwei Li; Shaolin Li; Hong Shan; Jian Lv; Tongtong Zhao; Junli Xia; Qihua Long; Sharon Steinberger; Adam Jacobi; Timothy Deyer; Marta Luksza; Fang Liu; Brent P. Little; Zahi A. Fayad; Yang Yang

Collaborative effort in data collection may facilitate improving the AI model. Difficulties on model training also arise due to the limited sample size. In this work we used a pre-trained TB model to select key slices to represent a full 3D CT scan. This approach can reduce computation of training a 3D convolution neural network, with a trade off on missing information in the slices that are not Another limitation is the bias towards SARS-CoV-2 patients in the training data, which, given the non-specific nature of the ground glass opacity and other features on chest CT images, potentially limits the usefulness of the current AI model to distinguish SARS-CoV-2 from other causes of respiratory failure. Therefore, our algorithm may be helpful in places with current high rates of COVID-19 disease, but is unlikely to provide as much usefulness in places or times where SARS-CoV-2 prevalence is low.