key: cord-0984979-ggwh91cc authors: Yang, Z.; Lin, D.; Chen, X.; Qiu, J.; Li, S.; Huang, R.; Sun, H.; Liao, Y.; Xiao, J.; Tang, Y.; Liu, G.; Wu, R.; Dai, Z. title: Distinguishing COVID-19 from influenza pneumonia in the early stage through CT imaging and clinical features date: 2020-04-22 journal: nan DOI: 10.1101/2020.04.17.20061242 sha: 2fbb5878c88923eace735df58b23329cc816d497 doc_id: 984979 cord_uid: ggwh91cc Purpose: To identify differences in CT imaging and clinical features between COVID-19 and influenza pneumonia in the early stage, and to identify the most valuable features in the differential diagnosis. Materials and Method: A consecutive cohort of 73 COVID-19 and 48 influenza pneumonia patients were retrospectively recruited from five independent institutions. The courses of both diseases were confirmed to be in the early stages (mean 2.66 (SD 2.62) days for COVID-19 and mean 2.19 (SD 2.10) days for influenza pneumonia after onset). The chi-square test, student`s t-test, and Kruskal-Wallis H-test were performed to compare CT imaging and clinical features between the two groups. Spearman or Kendall correlation tests between feature metrics and diagnosis outcomes were also assessed. The diagnostic performance of each feature in differentiating COVID-19 from influenza pneumonia was evaluated with univariate analysis. The corresponding area under the curve (AUC), accuracy, specificity, sensitivity and threshold were reported. Results: The ground-glass opacification (GGO) was the most common imaging feature in COVID-19, including pure-GGO (75.3%) and mixed-GGO (78.1%), mainly in peripheral distribution. For clinical features, most COVID-19 patients presented normal white blood cell (WBC) count (89.04%) and neutrophil count (84.93%). Twenty imaging features and 6 clinical features were identified to be significantly different between the two diseases. The diagnosis outcomes correlated significantly with the WBC count (r=-0.526, P<0.001) and neutrophil count (r=-0.500, P<0.001). Four CT imaging features had absolute correlations coefficients higher than 0.300 (P<0.001), including crazy-paving pattern, mixed-GGO in peripheral area, pleural effusions, and consolidation. Conclusions: Among a total of 1537 lesions and 62 imaging and clinical features, 26 features were demonstrated to be significantly different between COVID-19 and influenza pneumonia. The crazy-paving pattern was recognized as the most powerful imaging feature for the differential diagnosis in the early stage, while WBC count yielded the highest diagnostic efficacy in clinical manifestations. The coronavirus disease 2019 (COVID- 19) pandemic is a global crisis, which has killed more than seventy thousand peoples as of April 7, 2020 (1) . A clear picture of imaging and clinical manifestations of COVID-19 remains unknown. These manifestations of COVID-19 are protean and usually overlap with those of other viral pneumonia (2, 3) . In the early stage of COVID-19, the main radiological finding is the ground-glass opacity (GGO), especially the pure ground-glass opacity (2) , in the subpleural region, located unilaterally or bilaterally in the lower lobes (3) . The lesions can develop one or more lobes, with a slight preference for the lower right lobe (4) . However, these CT imaging findings are similar to those of influenza pneumonia (5, 6) . The main clinical manifestations of COVID-19, including fever, dry cough, and fatigue, are also non-specific (7, 8) . Both COVID-19 and influenza pneumonia are highly contagious and present similar symptoms. The US CDC has reported that some COVID-19 deaths have been miscategorized as influenza (9) . Unlike for influenza, no vaccine or antiviral agents are available for COVID-19 at the moment (10) . Moreover, the mortality rate for COVID-19 appears to be substantially higher than for influenza, about 5.6% vs 0.1% based on the primary data (1) . Therefore, the discrimination between COVID-19 and influenza is critical in clinical practice. Accurate imaging and clinical feature recognition can aid in early diagnosis for COVID-19 and thus prevent spreading and speed up treatment. In our previous study, we demonstrated that based on CT imaging and clinical manifestations alone, the pneumonia patients with and without COVID-19 can be distinguished (11) . Harrison et al. examined the performance of seven radiologists in differentiating COVID-19 from viral pneumonia . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. . https://doi.org/10.1101/2020.04. 17.20061242 doi: medRxiv preprint on chest CT results and found the average sensitivity of 80% and specificity of 84% (12) . However, we realized that about 44% of the viral pneumonia cases were Human Rhinovirus, and influenza pneumonia only accounted for about 15%. To our knowledge, no study has explored the differences between COVID-19 and influenza using CT imaging and clinical features. In this study, we aim to identify differences in CT imaging and clinical features between COVID-19 and influenza pneumonia in the early stage, and to identify the most valuable features in distinguishing COVID-19 from influenza pneumonia, based on multi-center data. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020 Non-contrast-enhanced chest CT imaging data were obtained from multiple hospitals of varied CT systems, including GE CT Discovery 750 HD (General Electric, US), SCENARIA 64 CT (Hitachi Medical, Japan), PHILIPS Ingenuity CT (PHILIPS, Netherlands), and Siemens SOMATOM Definition AS (Siemens, Germany) systems. All images were reconstructed into 1 mm slices with a slice interval of 0.8 mm. The detailed acquisition parameters are summarized in the supplementary material ( Table E1) . The baseline clinical data including course of disease, age, gender, body temperature, clinical symptoms (including cough, fatigue, sore throat, stuffy, and runny nose), total white blood cell (WBC) count, lymphocyte count, lymphocyte ratio, neutrophil count, neutrophil ratio and c-reactive protein (CRP) level were collected. According to the normal range used at individual hospital, the threshold value for WBC count, lymphocyte count, lymphocyte ratio, neutrophil count, neutrophil ratio and CRP level was set to 3.5~9.5×10 9 /L, 1.1~3.2×10 9 /L, 20.0~50.0%,1.8~6.3 ×10 9 /L,40.0~75.0% and 0.0~6.0 mg/L, respectively. A total of 26 quantitative and 22 qualitative imaging features were extracted for analysis. The descriptions of the CT imaging features are listed in the supplementary material (Table E2) . For the . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 22, 2020. . https://doi.org/10.1101/2020.04. 17.20061242 doi: medRxiv preprint extraction of CT qualitative and quantitative imaging features, two senior radiologists (Z.Y. and X.C., more than 15 years of experience) reached a consensus and were blinded to the clinical and laboratory findings. Lesion in the outer third of the lung was defined as peripheral and lesion in the inner two thirds of the lung was defined as central. The classification of the lesion size is based on a previous study (14) . The progression of lesion within each lung lobe was evaluated by scoring each lobe from 0 to 4 (15), corresponding to normal, 1% ~25% infection, 26%~ 50% infection, 51%~ 75% infection and more than 75% infection, respectively. The scores were combined for all 5 lobes to provide a total score ranging from 0 to 20. Figure 2 was one example of the evaluation of chest CT images. The CT imaging and clinical features were compared between COVID-19 and influenza pneumonia group by using the chi-square test (for nominal variable), the Kruskal-Wallis H test (for ordinal variable), or the student's t test (for continuous variable). The features with a significant difference between the two groups were extracted. Spearman or Kendall correlation test between feature metrics and diagnosis outcomes (i.e., 1 for COVID-19 and 0 for influenza pneumonia) were assessed for each extracted feature. The diagnostic performance of clinical and CT features in differentiating COVID-19 from influenza pneumonia was evaluated with univariate analysis. Additionally, corresponding area under the curve (AUC), accuracy, specificity, sensitivity and threshold were calculated. All statistical analyses for this study were performed with R (version 3.6.4, http: //www.r-project.org/). A two-tailed P-value < 0.05 indicated statistical significance. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 22, 2020 121 patients, including 73 COVID-19 and 48 influenza pneumonia were recruited in this study. The courses of both diseases were confirmed to be in the early stages, which were 2.66 ± 2.62 days for COVID-19 and 2.19 ± 2.10 days for influenza pneumonia after onset. A total of 15 clinical features of COVID-19 and influenza pneumonia patients are shown in Table 1 . Compare to COVID-19 patients, influenza pneumonia patients have higher temperature (P < 0.001), WBC count (P < 0.001), neutrophil count (P < 0.001), neutrophil rate (P=0.017), CRP level (P=0.033) and have lower lymphocyte rate (P=0.005). There is no significant difference in sex, age, cough, fatigue, sore throat, stuffy, runny nose, and lymphocyte count between the two groups. As shown in Figure 3 , most COVID-19 patients present normal WBC count (89.04%), neutrophil count (84.93%) and neutrophil rate (63.01%). A total of 1537 lesions were identified, with 1073 from COVID-19 group and 464 from influenza pneumonia group. The differences between COVID-19 and influenza pneumonia patients for CT quantitative and qualitative imaging features were showed in supplementary materials Table E3 and Table E4 , respectively. Those features with significant differences were presented in Table 2 . For imaging manifestations, 9 patients in the COVID-19 group (12.33%) and 3 patients in the influenza pneumonia group (6.25%) showed normal chest CT. Of all quantitative imaging features, COVID-19 patients have a greater total number of pure GGO (P = 0.01), total number of pure GGO in peripheral area (P = 0.003), total number of mixed GGO in peripheral area (P=0.016), and total number of . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. The correlation analysis and diagnostic performance of clinical features in distinguishing COVID-19 from influenza pneumonia were shown in Table 3 . The diagnosis outcomes correlated significantly with the WBC count (Spearman's r correlation, r = -0.526, P < 0.001) and neutrophil count (r = -0.500, P < 0.001). Lymphocyte rate and temperature have a weaker correlation with distinguishing COVID-19 from influenza pneumonia, with r = 0.310 (P < 0.001) and r = -0.433 (P < 0.001), respectively. However, little correlations were found for C-reactive protein and for neutrophil ratio in differential diagnosis. The WBC count yield a maximum AUC of 0.811 (95% CI: 0.731 ~ 0.890), follow by neutrophil count with the AUC of 0.795 (95% CI: 0.711 ~ 0.879). The distribution of WBC . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. . https://doi.org/10.1101/2020.04. 17.20061242 doi: medRxiv preprint count and neutrophil count in both groups were shown in Figure 4 . The correlation analysis and diagnostic performance of CT features in distinguishing COVID-19 from influenza pneumonia were shown in Table 4 . In COVID-19 diagnosis, the crazy paving pattern achieved the highest correlation of 0.379 (P < 0.001), which had an AUC of 0.687 (95% CI: 0.611 ~ 0.764). Mixed GGO in peripheral area had a correlation of 0.320 (P < 0.001). The consolidation and pleural effusions were more common in influenza pneumonia compared to COVID-19. The correlations for consolidation and for pleural effusions were -0.335 (P < 0.001) and -0.370 (P < 0.001), respectively. The typical CT imaging features of both diseases were illustrated in Figure 5 . . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. In this study, we compared CT imaging and clinical manifestations between COVID-19 and influenza pneumonia and identified the most valuable features for differential diagnosis. Among a total of 62 features, 20 imaging features and 6 clinical features were found to be significantly different. Correlation analysis showed that the WBC count had the highest correlation (r = -0.526, P < 0.001), with a threshold of 6.435 × 10 9 /L, followed by neutrophil count (r = -0.500, P < 0.001). Four CT imaging features were identified as the most significant for differential diagnosis in the early stage of both diseases, including crazy paving pattern, mixed ground-glass opacity in peripheral area, pleural effusions, and consolidation. The GGO in the periphery has become a recognized indicator of COVID-19 in the early stage (5, 16, 17) . In line with previous studies, we found that in the early stage of COVID-19, about 78% of patients had mixed GGO. However, this feature only ranked the third among the 26 extracted features for distinguishing COVID-19 from influenza. Crazy paving pattern, which also been reported in previous studies (3, 18, 19) , was considered to be the most powerful feature for the differential diagnosis. These two features were also reported in other coronavirus diseases, such as severe acute respiratory syndrome (SARS) and middle east respiratory syndrome (MERS) (20, 21) . The pathology of COVID-19 was confirmed to greatly resemble those of SARS and MERS (22, 23) . Tian et al. reported that the lungs of COVID-19 patients exhibited edema, proteinaceous exudate, focal reactive hyperplasia of pneumocytes with patchy inflammatory cellular infiltration, and multinucleated giant cells (24) , which can cause the thickening of interlobular septa, and represented as crazy paving pattern. Consistent with previous reports (25) , the pleural effusions are very rare in COVID-19 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (28), which was also confirmed by Shi et al. (5) . Therefore, in the follow up of the disease, the difference of this feature between the two diseases may be weakened. The bronchial wall thickening was proved to be not significantly different between influenza and COVID-19 pneumonia (p = 0.715), which indicated that both diseases could affect airway walls. Recently, RT-PCR and serological antibody tests are widely adopted for COVID-19 diagnosis. However, false-negative cases using RT-PCR have been reported in several studies (29-31). Serum antibody test was shown to have good performance for the diagnosis of COVID-19, with sensitivity of 88.66% and specificity of 90.63% (32). Because it likely takes the body one to three weeks to produce the antibodies, antibody test is unable to diagnose the illness in the early stage. To et al (33) found that IgG or IgM antibody increased for most patients at 10 days or later after symptom onset. Therefore, imaging and clinical findings have the advantage to reflect the disease earlier. To our best knowledge, our study is the first to evaluate the significant statistical difference of CT imaging and clinical . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. . https://doi.org/10.1101/2020.04. 17.20061242 doi: medRxiv preprint features between COVID-19 and influenza pneumonia. It is worth noting that every individual feature has limited diagnostic efficacy and thus the combination of multiple features will be the trend of future research. There are several limitations in this study. First, in order to evaluate the differential diagnosis in the early stage, we only compare the initial CT scanning both in COVID-19 and influenza pneumonia. Since the CT manifestations change with the course of the disease (34), our results may have a bias at different time windows. Second, there may be some inherent deviations in the multi-center retrospective design (35), since the scanning protocols are slightly diverse in different hospitals. Finally, although the preliminary results are promising, further validation on a larger and independent dataset is needed to determine the potential of these features for distinguishing COVID-19 from influenza pneumonia. After validation, further diagnostic models may be created based on these features. In conclusion, a total of 1537 lesions and 62 features were compared between COVID-19 and influenza pneumonia patients. Twenty-six features were significantly different between the two groups. In CT imaging, the crazy paving pattern was recognized as the most powerful feature in the differential diagnosis in the early stage, with AUC of 0.687 (95% CI: 0.611~0.764). In clinical manifestations, white blood cell count had the highest AUC of 0.811 (95% CI: 0.731~0.890). These findings help to distinguish COVID-19 from influenza pneumonia. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. . https://doi.org/10.1101/2020.04. 17.20061242 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. . https://doi.org/10.1101/2020.04. 17.20061242 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. . . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted April 22, 2020 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 22, 2020 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 22, 2020. . https://doi.org/10.1101/2020.04. 17.20061242 doi: medRxiv preprint Novel Coronavirus (2019-nCoV) situation reports Temporal Changes of CT Findings in 90 Patients with COVID-19 Pneumonia: A Longitudinal Study Time Course of Lung Changes On Chest CT During Recovery From 2019 Novel Coronavirus (COVID-19) Pneumonia Early Clinical and CT Manifestations of Coronavirus Disease 2019 (COVID-19) Pneumonia Radiological findings from 81 patients with COVID-19 pneumonia in Wuhan, China: a descriptive study. The Lancet Infectious Diseases Comparison of patients hospitalized with influenza A subtypes H7N9, H5N1, and 2009 pandemic H1N1. Clinical infectious diseases : an official publication of the Infectious Diseases Society of America Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China. The Lancet A novel coronavirus from patients with pneumonia in China US CDC Director Robert Redfield Admitted that Coronavirus Deaths Have Preventing a covid-19 pandemic A diagnostic model for coronavirus disease 2019 (COVID-19) based on radiological semantic and clinical features: a multi-center study Performance of radiologists in differentiating COVID-19 from viral pneumonia on chest CT Technical guidelines for seasonal influenza vaccination in China CT correlation with outcomes in 15 patients with acute Middle East respiratory syndrome coronavirus CT imaging features of 2019 novel coronavirus (2019-nCoV) Chest CT Features of COVID-19 in Chest CT Findings in 2019 Novel Coronavirus (2019-nCoV) Infections from Wuhan, China: Key Points for the Radiologist COVID-19): A Perspective from China Imaging and clinical features of patients with 2019 novel coronavirus SARS-CoV-2 Middle East Respiratory Syndrome Coronavirus (MERS-CoV) Infection: Chest CT Findings High-Resolution CT Findings in Patients with Severe Acute Respiratory Syndrome: A Pattern-Based Approach Pathological findings of COVID-19 associated with acute respiratory distress syndrome. The Lancet Respiratory Medicine Three Emerging Coronaviruses in Two Decades: The Story of SARS, MERS, and Now COVID-19 Pulmonary Pathology of Early-Phase 2019 Novel Coronavirus (COVID-19) Pneumonia in Two Patients With Lung Cancer Chest CT manifestations of new coronavirus disease 2019 (COVID-19): a pictorial review