Clinical characteristics, CT findings and AI application in pregnant women with COVID-19 pneumonia

Objective: To analyze the clinical and radiological characteristics of COVID-19 patients in pregnancy, and to study the value of artificial intelligence (AI) in the diagnosis of COVID-19. Methods: This study retrospectively included 70 female COVID-19 patients with CT images and complete clinical data, and the range of ages were 22-39 years old. The mild COVID-19 patients with normal CT findings were excluded. All patients underwent AI analysis. Of them, 30 patients were in pregnancy (age, 29.4 ± 4.7 years), the other 40 COVID-19 patients (age, 29.6 ± 3.9 years) were not in pregnancy (ordinary group). The CT characteristics (lesions, distribution, involvement of lung lobes, and accompanying signs) were recorded for each patient. Chi-square test and t test or Wilcoxon test were used to compare the clinical features and CT characteristics between the two groups. Kappa analysis was used to analyze the consistency of identifying CT characteristics between AI and radiologist. Results: The clinical manifestations in the two groups were common type. The pregnant group presented more mild symptoms than the ordinary group: No symptoms, 21 cases, 70% vs 4 cases, 10% and less cases with fever and other symptoms (oppression or pain in the chest, fatigue, etc.) (P < 0.05). Lymphocyte percentage and neutrophil granulocyte rate, D-dimer and C-reactive protein were higher in the pregnant group than in the ordinary group (P < 0.05). Based on CT appearances, the pregnant group was mostly in early stage, and the ordinary group was mostly in progressive stage (P < 0.05). Comparing the CT characteristics of the patients in the two groups, there was more single ground-glass opacity (GGO) in the pregnant group and multiple GGO in the ordinary group (P < 0.05), the other CT features did not statistically differ between the two groups (P > 0.05). CT characteristics identified by AI and radiologist agreed well (κ > 0.8), the diagnostic consistency of the identifying peripheral area and GGO were general (κ = 0.41-0.80), while the consistency of identifying fibrous stripes was poor (κ = 0.268). Conclusions: COVID-19 patients in pregencymostlyhaveno obvious symptoms, and their CT characteristics and laboratory tests are different from those with general COVID-19 patients. AI couldbe used as an assistant tool in the diagnosis of COVID-19. (authors)