Analysis of clinical features and imaging signs of COVID-19 with the assistance of artificial intelligence

OBJECTIVE: To explore the CT imaging features/signs of patients with different clinical types of Coronavirus Disease 2019 (COVID-19) via the application of artificial intelligence (AI), thus improving the understanding of COVID-19. PANTIENTS AND METHODS: Clinical data and chest CT imaging features of 58 patients confirmed with COVID-19 in the Fifth Medical Center of PLA General Hospital were retrospectively analyzed. According to the Guidelines on Novel Coronavirus-Infected Pneumonia Diagnosis and Treatment (Provisional 6th Edition), COVID-19 patients were divided into mild type (7), common type (34), severe type (7) and critical type (10 patients). The CT imaging features of the patients with different clinical types of COVID-19 types were analyzed, and the volume percentage of pneumonia lesions with respect to the lung lobes (where the lesion was located) and to the whole lung was calculated with the use of Al software. SPSS 21.0 software was used for statistical analysis. RESULTS: Common clinical manifestations of COVID-19 patients: fever was found in 47 patients (81.0%), cough in 31 (53.4%) and weakness in 10 (17.2%). Laboratory examinations: normal or decreased white blood cell (WBC) counts were observed in 52 patients (89.7%), decreased lymphocyte counts (LCs) in 14 (24.1%) and increased C-reactive protein (CRP) levels in 18 (31.0%). CT imaging features: there were 48 patients (94.1%) with lesions distributed in both lungs and 46 patients (90.2%) had lesions most visible in the lower lungs; the primary manifestations in patients with common type COVID-19 were ground -glass opacities (GG0s) (23/34, 67.6%) or mixed type (17/34, 50.0%), with lesions mainly distributed in the periphery of the lungs (28/34, 82.4%); the primary manifestations of patients with severe/critical type COVID-19 were consolidations (13/17, 76.5%) or mixed type (14/17, 82.4%), with lesions distributed in both the peripheral and central areas of lungs (14/17,82.4%); other common signs, including pleural parallel signs, halo signs, vascular thick ening signs, crazy -paving signs and air bronchogram signs, were visible in patients with different clinical types, and pleural effusion was found in 5 patients with severekritical COVID-19. Al software was used to calculate the volume percentages of pneumonia lesions with respect to the lung lobes (where the lesion was located) and to the whole lung. There were significant differences in the volume percentages of pneumonia lesions for the superior lobe of the left lung, the inferior lobe of the left lung, the superior lobe of the right lung, the inferior lobe of the right lung and the whole lung among patients with different clinical types (p<0.05). The area under the ROC curve (AUC) of the volume percentage of pneumonia lesions for the whole lung for the diagnosis of severe/critical type COVID-19 was 0.740, with sensitivity and specificity of 91.2% and 58.8%, respectively. CONCLUSIONS: The clinical and CT imaging features of COVID-19 patients were characteristic to a certain degree; thus, the clinical course and severity of COVID-19 could be evaluated with a combination of an analysis of clinical features and CT imaging features and assistant diagnosis by AI software.