Computer Aided Diagnosis Research For Severity Assessment Of COVID-19 With CT Images

Millions of positive COVID-19 patients are suffering from the pandemic around the world since the outbreak.And as more and more patients enter hospitals for treatment,a critical step in the hierarchical treatment and treatment tracking is severity assessment,which is quite challenging with the limited medical resources.Currently,several artificial intelligence algorithms have been developed for the severity assessment of COVID-19.However,imprecise severity assessment and insufficient data are still obstacles.To address these issues,we proposed a novel preprocessing method of 3D CT images and two artificial intelligence algorithms integrating prior knowledges.The details of these two works are as follows:1.We proposed a preprocessing method of 3D CT images.We firstly used lung and lesion segmentation algorithms to obtain lesion area in the left and right lungs of patients’ CT images,then put the lesion area into a small cube and extracted multi-view lesion slices based on the cube.On the basis of retaining rich 3D information,the preprocessing method in this paper greatly reduces the amount of 3D input information and the complexity of algorithm.Moreover,it effectively converts 3D images to multiple 2D lesion slices and alleviates the few-shot issue.2.We proposed two artificial intelligence algorithms integrating prior knowledges（a traditional machine-learning-based severity assessment algorithm,a deep-learningbased severity assessment algorithm）.We integrated four basic clinical metadata（sex,age,time of onset,time interval）and two adjacent CT scans as prior knowledges for improving the performance of artificial intelligence algorithms.We evaluated the proposed methods on 449 COVID-19 patients who underwent continuous chest CT examinations,two algorithms achieved accuracy of 85.16% and86.72%,respectively.The results indicated our methods achieve good performance,and may contribute a potential solution to severity assessment of COVID-19 patients,by utilizing the proposed preceseeing method and integrating prior knowledge.