Deep Learning-Based Quantitative Computed Tomography Model in Predicting the Severity of COVID-19: A Retrospective Study in 196 Patients

Abstract

Background: The assessment of the severity of Corona Virus Disease 2019 (COVID-19) by clinical presentations cannot met the urgently clinical need so far. We aimed to establish a model of deep learning (DL)-based quantitative computed tomography (CT) and initial clinical features in prediction of the severity of COVID-19.

Methods: One hundred and ninety-six hospitalized patients with COVID-19 were enrolled from Jan 20 to Feb 10, 2020 in our center, and were divided into the severe and non-severe group. The clinico-radiological data at admission were compared between the two groups. A least absolute shrinkage and selection operator (LASSO) logistic regression model for predicting the severity of COVID-19 was established, and the area under the receiver operating characteristic (AUC) values of the model, quantitative CT parameters significantly different in univariate analysis, and pneumonia severity index (PSI) were compared.

Findings: In comparison with non-severe group (151 cases), severe patients (45 cases) had higher PSI (p<0.001). DL-based quantitative CT indicated that the mass of infection (MOI_CT) and the percentage of infection (POI_CT) in the severe group was higher (p<0.001).The LASSO logistic regression model was based on MOI_CT together with clinical features including age, lactate dehydrogenase (LDH), C-reactive protein (CRP), CD4⁺ T cell counts. The AUC value for the model was 0.890, and was significantly higher than that for MOI_CT, POI_CT, or PSI score (all p<0.01).

Interpretation: The DL-based quantitative CT model has more efficiency in predicting the severity of COVID-19 than quantitative CT parameters and PSI score do.

Funding Statement: The authors stated: "None."

Declaration of Interests: The authors declare no competing interests.

Ethics Approval Statement: The Institutional Review Board of Shanghai Public Health Clinical Center, Fudan University approved the study protocol. Informed consent was waived because of the retrospective nature of the study.

Keywords: Corona Virus Disease 2019, Computed tomography, Deep learning, Multivariate analysis, Predicting