key: cord-0835445-zejil78m
authors: Zhang, Bo; Wang, Xia; Tian, Xiaoyan; Zhao, Xiaoying; Liu, Bin; Wu, Xingwang; Du, Yaqing; Huang, Guoquan; Zhang, Qing
title: Differences and prediction of imaging characteristics of COVID-19 and non-COVID-19 viral pneumonia: A multicenter study
date: 2020-10-16
journal: Medicine (Baltimore)
DOI: 10.1097/md.0000000000022747
sha: 5b7760a5ba59e18a284053e49ade18656962d537
doc_id: 835445
cord_uid: zejil78m

To study the differences in imaging characteristics and prediction of COVID-19 and non-COVID-19 viral pneumonia through chest CT. Chest CT data of 128 cases of COVID-19 and 47 cases of non-COVID-19 viral pneumonia confirmed by several hospitals were retrospectively collected, the imaging performance was evaluated and recorded, different imaging features were statistically analyzed, and a prediction model and independent predicted imaging features were obtained by multivariable analysis. COVID-19 was more likely than non-COVID-19 pneumonia to have a high-grade ground glass opacities (P = .01), extensive lesion distribution (P < .001), mixed lesions of varying sizes (27.7% vs 57.0%, P = .001), subpleural prominence (23.4% vs 86.7%, P < .001), and lower lobe prominence (48.9% vs 82.0%, P < .001). However, peribronchial interstitial thickening was more likely to occur in non-COVID-19 viral pneumonia (36.2% vs 19.5%, P = .022). The statistically significant differences from multivariable analysis were the degree of ground glass opacities (P = .001), lesion distribution (P = .045), lesion size (P = .020), subpleural prominence (P < .001), and lower lobe prominence (P = .041). The sensitivity and specificity of the model were 94.5% and 76.6%, respectively, with an AUC of 0.91. The imaging characteristics of COVID-19 and non-COVID-19 viral pneumonia are different, and the prediction model can further improve the specificity of chest CT diagnosis.

By the end of December 2019, novel coronavirus 2019 pneumonia (COVID-19) began in Wuhan, Hubei, China, and spread rapidly on a large scale worldwide. By the beginning of March 2020, COVID-19 had reached 114 countries, affected more than 100,000 people and killed more than 4000 people. On March 12, 2020, the World Health Organization defined COVID-19 as a pandemic. [1] Although the virus causing COVID-19, SARS-COV-2, belongs to the same genus as SARS-COV and MERS-COV, its transmissibility and case fatality rate are quite different. The binding capacity of the surface S protein on SARS-COV-2 to angiotensinase 2 is approximately 10 to 20 times that of SARS-COV, [2] so its infectivity is significantly higher than that of SARS-COV. The case fatality rate of COVID-19 (3%-4%) is lower than that of SARS (10%) and MERS (approximately 36%), but it is significantly higher than that of H1N1 virus (0.1%-1%). [3] [4] [5] The early symptoms of the disease are mainly fever and cough, and some patients progress to severe acute respiratory distress syndrome and respiratory failure, as well as different degrees of involvement of systemic organs. [6, 7] Early detection, early isolation and timely treatment are the key points for the prevention and treatment of the epidemic. RT-PCR and chest CT in the early period are the main means for the early detection of the disease. The sensitivity of RT-PCR was reported to be lower than that of chest CT. [8] Due to the sampling method, supply shortage and other reasons, it is difficult for the first RT-PCR to achieve timely and accurate diagnosis in a large number of suspected patients, which leads to a large number of missed diagnoses and increases the risk of transmission. COVID-19 overlaps with other non-COVID-19 viral pneumonias in terms of imaging findings. In addition, when the patients exposure history is unclear, it is difficult for diagnostic imaging doctors to distinguish the 2. Accurate diagnosis requires subjective experience, and failure to make a timely or accurate diagnosis increases the risk of epidemic spread or excessive medical treatment. Nonviral pneumonia is easy to differentiate on imaging, and coupled with laboratory examination, clinician diagnosis is not difficult. In this study, we analyzed the characteristics of COVID-19 and other viral pneumonias on chest CT for the first time and made a COVID-19 prediction model based on these analyses. This method can effectively exclude subjective factors to make predictions and distinctions, and especially when the patient's contact history is unclear, it can quantify the incidence probability of COVID-19 relative to that of other viral pneumonias to improve the specificity of chest CT.

The clinical history data of 132 cases of COVID-19 diagnosed in several hospitals from January to February 2020 and 76 cases of non-COVID-19 viral pneumonia diagnosed clinically from January 2016 to December 2019 were retrospectively collected. Four CT-negative COVID-19 cases and 29 cases with other viral pneumonia without a nucleic acid test or viral antibody test were excluded. The final number of cases included was 128 COVID-19 and 47 non-COVID-19 viral pneumonias. The first chest CT images of all patients after symptom onset were collected and recorded. Each hospitals ethics committee approved the study.

All patients were scanned by 2 scanners: a 64-MDCT Light Speed VCT (GE healthcare) and Somatom Emotion (Siemens healthcare). The collection parameters were set to 120 kVp, 100 to 200 mAs, spacing of 0.75 to 1.5, and collimating of 0.625 to 5 mm. All imaging data were reconstructed using the reconstruction algorithm of medium sharpness, and the thickness of the layers was 0.625 to 5 mm. CT images were obtained in the supine position with complete inspiration.

Two radiologists with 10 years of radiographic experience independently assessed the images without prior knowledge of the clinical diagnosis. Differences between the 2 radiologists were resolved by consensus to achieve consistent results. All images were evaluated at the axial position under the lung window (width, 1600 HU; level, -600 HU) and the mediastinal window (width, 400 HU; level, 40 HU). The following imaging features were recorded:

1. consolidation degree and ground glass opacity degree (grade 0: less than 10% of all lesions; grade 1 accounted for 30% of all lesions; grade 2 accounted for 30%-60% of all lesions; grade 3 accounted for 60%-90% of all lesions; and grade 4 accounted for greater than 90% of all lesions); 2. main lesion morphology (nodular, patchy, or large patchy); 3. lesion distribution (localized or small, scattered, multiple, or diffuse); 4. lesion size (small lesion less than 3 cm, large lesion more than 3 cm, or mixed lesions); 5. air bronchogram; 6. fibrotic streaks; 7. subpleural prominence (lesions located mainly in the subpleural space); 8. distribution along the bronchovascular bundle; 9. interlobular septal or reticular thickening; 10. peribronchovascular interstitial thickening (CT demonstrates vascular thickening or bronchial wall thickening); 11. pleural thickening; 12. lower lobe prominence (lesions located mainly in the lower lobe); 13. total number of involved lung lobes; and 13. distribution of lesions in different lung lobes.

SPSS statistical software (version 25.0; SPSS Inc., Chicago, Illinois, United States) was used for analysis. Continuous variables are expressed as averages ± standard deviations and compared by the Mann-Whitney U test. Categorical variables are expressed as numbers and percentages and are compared by the Chi-Squared test or Fishers exact test. P < .05 was considered statistically significant. The results of univariate analysis (P < .05) and clinically significant factors (P > .05) were included in the multivariable analysis. Multivariable analysis and prediction models were established by binary logistic regression. A receiver operating characteristic (ROC) curve was used to evaluate the performance of the prediction model.

The distribution of non-COVID-19 viral pneumonia pathogens is shown in Figure 1 , including 16 cases of influenza A, 9 cases of cytomegalovirus (CMV), 8 cases of Epstein-Barr virus (EBV), 5 cases of adenovirus, 4 cases of influenza B, 3 cases of human parainfluenza virus (HPIV) and 2 cases of varicella-zoster virus (VZV). COVID-19 patients included 79 males and 49 females, with an average age of 46.1 ± 12.8 years. Twenty one males and 26 females were included in the non-COVID-19 viral pneumonia group, with an average age of 41.9 ± 23.1 years. Men were more common in the COVID-19 group than in the non-COVID-19 group (61.7%, P = .045). The average time between symptom onset and initial CT examination was 7.0 ± 4.3 days for non-COVID-19 viral pneumonia and 6.0 ± 3.4 days for COVID-19. The main clinical features of the 2 groups are summarized in Table 1 . The main symptoms were fever, cough, sore throat, fatigue, and muscle soreness.

The imaging characteristics of non-COVID-19 viral pneumonia and COVID-19 are shown in Table 2 ). Compared with non-COVID-19 viral pneumonia, COVID-19 was more likely to have high-grade ground glass opacities (P = .010) (Fig. 2AD) , a wide distribution of lesions (P < .001), mixed lesions of different sizes (27.7% vs 57.0%, P = .001) (Fig. 2CE) , subpleural prominence (23.4% vs 86.7%, P < .001) (Fig. 2ABCE) , and lower lobe prominence (48.9% vs 82.0%, P < .001) (Fig. 2FG) . However, peribronchial interstitial thickening was more likely to occur in non-COVID-19 viral pneumonia (36.2% vs 19.5%, P = .022) (Fig. 2CGH) . The distribution of lesions in different lung lobes is shown in Figure 3 .

In both COVID-19 and non-COVID-19 viral pneumonia, the right lower lobe (83.6% and 76.6%) was more involved than the other lobes. The left upper lobe was more involved in COVID-19 than in non-COVID-19 viral pneumonia (74.2% vs 57.4%, P = .032). After adjusting for confounding factors by binary logistic regression analysis, independent predicted factors were identified (Table 3) , including the degree of ground glass opacities (P = .001), lesion distribution (P = .045), lesion size (P = .020), subpleural prominence (P < .001), and lower lobe prominence (P = .041). The prediction model established by logistic regression was statistically significant; the Chi-Squared value was 94.016, P < .001; the area under the ROC curve (AUC) (Fig. 4 ) was 0.91; the specificity was 76.6%; the sensitivity was 94.5%; and the accuracy was 89.7%.

COVID-19, a highly infectious and high-mortality infectious disease, has become a global epidemic. Early diagnosis and recognition of the disease are the keys to controlling the epidemic. Although the diagnosis of COVID-19 depends on RT-PCR, the sensitivity of the first RT-PCR of sputum and throat swab tests is approximately 50%-to 70%, while that of chest CT is over 90%. [8] [9] [10] As a noninvasive examination method, chest CT can quickly and conveniently screen suspected cases and act as an important complementary examination method for RT-PCR in epidemic areas. However, the specificity of chest CT is only 25%, [8, 10, 11] which leads to the emergence of a large number of false positive patients, bringing a certain sense of panic to the patients and placing a certain burden on medical management. When the infection source exposure history is unclear, it is difficult to distinguish COVID-19 from other viral pneumonias, as there are many similarities in imaging manifestations, while bacterial pneumonia is easy to distinguish via laboratory blood cell examination. [12] By analyzing the imaging manifestations of COVID-19 and non-COVID-19 viral pneumonia, this study identified the relatively statistically significant imaging feature differences, calculated the independently predicted imaging features excluding confounding factors through a multivariable binary logistic regression equation, and established a relevant prediction model based on the factors included in the regression equation.

According to this multicenter study of 128 COVID-19 cases, we found that COVID-19 early imaging features include mainly an increased degree of ground glass opacities, main large patchy lesions, mixed lesions and multiple lesions, air bronchogram, fibrotic streaks, subpleural prominence, distribution along the bronchovascular bundle, interlobular septal or reticular thickening, peribronchovascular interstitial thickening, pleural thickening, and lower lobe prominence. These findings are broadly similar to those previously reported in the literature. However, interlobular septal thickening and bronchial interstitial thickening occurred less in our study than in previous reports, [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] which may be related to the fact that our case collection represents the first chest CT examination of the patients, and the pathological changes are still in the early stage, such as the imaging features of lobular septal thickening and bronchial and vascular thickening not appearing. Our study also found that compared with non-COVID-19 viral pneumonia patients, COVID-19 patients were more likely to have high-grade ground glass opacities, extensive lesion distribution, mixed lesions of different sizes, and subpleural and lower lobe prominence, while peribronchial interstitial thickening was more likely to occur in non-COVID-19 viral pneumonia patients. Although viral pneumonia is caused mainly by interstitial changes, different genera of viruses have different pathogeneses, leading to different imaging changes. [12] We also found that in COVID-19 and non-COVID-19 viral pneumonia patients, the right lower lobe was more often involved, which may be related to the short and thick physiological structure of the right lower lobe bronchus that may make it easier for the virus to enter this lobe. Based on multivariable analysis, the degree of ground glass opacities, lesion distribution, lesion size, subpleural prominence, and lower lobe prominence were found to be potential independent factors to predict the probability of COVID-19 versus non-COVID-19 viral pneumonia. The specificity, sensitivity and AUC of the prediction model were 76.6%, 94.5%, and 0.91%, respectively, indicating that the model had good diagnostic efficiency and classification function. Our study will enhance radiologists understanding of COVID-19 imaging findings and further enhance the specificity of chest CT.

Our study also has some limitations. The sample size of our retrospective collection of non-COVID-19 viral pneumonia was too small, which may include some common cases and lack certain representativeness. We did not assess the consistency of the judgments made by different radiologists using the model. Although our study attempted to predict the occurrence of COVID-19 versus non-COVID-19 viral pneumonia using a more quantitative and objective approach, the recognition of image characteristics varies among radiologists with different experiences.

In conclusion, the imaging characteristics of COVID-19 and non-COVID-19 viral pneumonia are different, and the prediction model can further improve the specificity of chest CT diagnosis. 

World Health Organization website WHO Director-General's opening remarks at the media briefing on COVID-19

Cryo-EM structure of the 2019-nCoV spike in the prefusion conformation

SARS case-fatality rates

Relation Between Chest CT Findings and Clinical Conditions of Coronavirus Disease (COVID-19) Pneumonia: A Multicenter Study

Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China

Pathological findings of COVID-19 associated with acute respiratory distress syndrome

Neurologic Manifestations of Hospitalized Patients With Coronavirus Disease

Sensitivity of Chest CT for COVID-19: Comparison to RT-PCR

Laboratory diagnosis and monitoring the viral shedding of nCoV infections

Correlation of Chest CT and RT-PCR Testing for Coronavirus Disease 2019 (COVID-19) in China: A Report of 1014 Cases

Time Course of Lung Changes at Chest CT during Recovery from Coronavirus Disease 2019 (COVID-19)

Radiographic and CT features of viral pneumonia

Chest CT Findings in Coronavirus Disease-19 (COVID-19): Relationship to Duration of Infection

The characteristics and clinical value of chest CT images of novel coronavirus pneumonia

Editor: Hyunjin Park.The authors report no conflicts of interest.The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

We thank Wendong Liu, Jinshun Yao, Xingming Sang, Bin Wang for their support in data collection and statistical consultation.