Visual classification of three computed tomography lung patterns to predict prognosis of COVID-19: a retrospective study

Abstract

Background: Quantitative evaluation of radiographic images has been developed and suggested for the diagnosis of coronavirus disease 2019 (COVID-19). However, there are limited opportunities to use these image-based diagnostic indices in clinical practice. Our aim in this study was to evaluate the utility of a novel visually-based classification of pulmonary findings from computed tomography (CT) images of COVID-19 patients with the following three patterns defined: peripheral, multifocal, and diffuse findings of pneumonia. We also evaluated the prognostic value of this classification to predict the severity of COVID-19.

Methods: This was a single-center retrospective cohort study of patients hospitalized with COVID-19 between January 1st and September 30^th, 2020, who presented with suspicious findings on CT lung images at admission (n = 69). We compared the association between the three predefined patterns (peripheral, multifocal, and diffuse), admission to the intensive care unit, tracheal intubation, and death. We tested quantitative CT analysis as an outcome predictor for COVID-19. Quantitative CT analysis was performed using a semi-automated method (Thoracic Volume Computer-Assisted Reading software, GE Health care, United States). Lungs were divided by Hounsfield unit intervals. Compromised lung (%CL) volume was the sum of poorly and non-aerated volumes (- 500, 100 HU). We collected patient clinical data, including demographic and clinical variables at the time of admission.

Results: Patients with a diffuse pattern were intubated more frequently and for a longer duration than patients with a peripheral or multifocal pattern. The following clinical variables were significantly different between the diffuse pattern and peripheral and multifocal groups: body temperature (p = 0.04), lymphocyte count (p = 0.01), neutrophil count (p = 0.02), c-reactive protein (p < 0.01), lactate dehydrogenase (p < 0.01), Krebs von den Lungen-6 antigen (p < 0.01), D-dimer (p < 0.01), and steroid (p = 0.01) and favipiravir (p = 0.03) administration.

Conclusions: Our simple visual assessment of CT images can predict the severity of illness, a resulting decrease in respiratory function, and the need for supplemental respiratory ventilation among patients with COVID-19.

Keywords: COVID-19; Computed tomography; Respiratory function; Retrospective study.

Fig. 1

Flowchart depicting the patient selection process. COVID-19: coronavirus disease 2019, CT: computed tomography

Fig. 2

Computed tomography (CT) images for the three patterns of COVID-19 pneumonia. The pulmonary opacities are classified into peripheral, multifocal, and diffuse patterns. a In the peripheral pattern, parenchymal opacification appears in the inner peripheral zone. b In the multifocal pattern, parenchymal opacification is apparent in the central and peripheral regions. c The diffuse pattern reveals generalized pulmonary involvement, with regional inhomogeneity. d–f Semi-automated segmentation using Thoracic VCAR software (GE Healthcare, USA). Blue areas represent normal lung parenchyma in the -501, -900 HU interval; light blue areas represent hyperinflated lung in the -901, -1000 HU; yellow areas represent poorly aerated lung in the -500, -100 HU interval; and red areas represent non-aerated lung in the 100, -100 HU interval. COVID-19, coronavirus disease 2019; HU, Hounsfield unit; VCAR, Volume Computer-Assisted Reading

Fig. 3

Violin plots for the three CT patterns and duration of intubation. The line represents 95% CI, the box represents the interquartile range, and the point is representative of the median. Density plot width indicates the frequency. There was a significant difference between the three CT patterns and duration of intubation (Kruskal–Wallis test, p < 0.0001). On the Mann–Whitney U test, there were significant differences between the peripheral and diffuse patterns and between the multifocal and diffuse patterns (p = 0.003 and p = 0.001, respectively). CT, computed tomography

Fig. 4

Violin plots for the three CT patterns and compromised lung (% CL). There was a significant difference between the three CT patterns and compromised lung (% CL) (Kruskal–Wallis test, p = 0.003). On the Mann–Whitney U test, there were significant differences between the peripheral and diffuse patterns and between the multifocal and diffuse patterns (p = 0.001 and p = 0.002, respectively). CT, computed tomography

Fig. 5

Results of the ROC curve analysis for the three patterns in predicting intubation. The three CT patterns predicted the tracheal intubation, with an area under the ROC curve of 0.77. CT, computed tomography; ROC, receiver operating characteristic