诊断病理学：开放获取

抽象的

Convolutional Neural Network-Based Image Recognition Systems: Detecting the Peripheral Granular Lymphocytopenia and Dysmorphic Leukocytosis as Prognostic Markers of COVID-19

Yuki Horiuchi, Yoko Tabe

Developing prognostic markers can aid in clinical decision making. Peripheral Blood (PB) testing is a simple and basic test that can be performed at any facility. Changes in blood cell morphology as prognostic indicators of coronavirus infection (COVID-19) have been studied using an automated image recognition system based on Convolutional Neural Networks (CNNs). The incidence of anemia, lymphopenia, and leukocytosis was significantly higher in severe cases than in mild cases. Granulocyte counts were persistently decreased in the lethal cases but remained normal or higher in the mild cases. A transient increase in granulocytic lymphocytes was associated with survival in patients with severe infection, and neutrophilic dysplasia was observed in severe COVID-19 cases. Giant neutrophil number and toxic granulation tissue/Döhle bodies were increased in severe cases. Erythrocyte distribution was significantly larger in severe cases than in mild cases. Blood cell calculation using basic PB testing and the detection of morphological abnormalities utilizing CNN may be useful in predicting the prognosis of COVID-19.