Predicting COVID-19 malignant progression with AI techniques

Xiang Bai; Cong Fang; Yu Zhou; Song Bai; Zaiyi Liu; Qianlan Chen; Yongchao Xu; Tian Xia; Shi Gong; Xudong Xie; Dejia Song; Ronghui Du; Chunhua Zhou; Chengyang Chen; Dianer Nie; Dandan Tu; Changzheng Zhang; Xiaowu Liu; Lixin Qin; Weiwei Chen

Our work is novel because it is the first study in which complementary data of quantitative CT sequence and clinical data is used to analyze the problem of COVID-19 malignant progression prediction. Experimental results show that both of them have a significant reference value for this problem and can obtain more accurate prediction results. Furthermore, there is very little literature to date modeling the spatial information of the quantitative CT data and considering the time-series information of patients. This information has important reference value for the prediction of patients with potential malignant progression. Specifically, the quantified spatial CT data is converted into a two-dimensional matrix format according to the real distribution of lung areas. And then, the spatial CT data was combined with the disease course and other infection sign type features. Finally, the hybrid CT data was folded into sequences following the sequence of clinical and patient-specific information.