Patients
We retrospectively enrolled 91 patients fulfilling the inclusion criteria: patients who underwent high-resolution CT within 7 days after the onset of symptoms and had the first consultation at the general hospital from Jan 10 to Feb 28, 2020. Of those 91 patients, 30 cases of COVID-19 were confirmed with WHO interim guidance, and 43 cases of other aetiology or clinically confirmed non-COVID-19 were finally included in our cohort. Suspected COVID-19 cases with abnormal chest CT findings (one COVID-19 and 1 non-COVID-19 with normal chest CT were excluded) were included with the following inclusion criteria[1]: (1) epidemiological exposure history within 14 days before the onset of symptoms – i) travel/residence history in Wuhan; ii) travel/residence history in Hubei but not Wuhan; iii) exposure history to confirmed cases or community, respiratory symptoms related patient; iv) cluster onset; (2) presented with fever and/or respiratory symptoms within 7 days of CT examination; and (3) normal or low white blood cell count and lymphocyte count at early onset. The exclusion criteria were as follows: (1) images with excessive motion artefact (one non-COVID-19 was excluded); (2) children and pregnant women (three COVID-19 and 9 non-COVID-19 were excluded); (3) lost to follow-up (three non-COVID-19)
The clinical data including age, sex, exposure history and laboratory parameters of all patients are summarized in Table 1.
Pathogenic evidence: a nucleic acid test by RT-PCR was used to detect the new coronavirus in respiratory samples. All enrolled patients had final diagnoses of twice-positive RT-PCR to confirm COVID-19, more than or equal to twice-negative RT-PCR (range 2-5 times) or at least one negative RT-PCR with other pathogens (mycoplasma pneumonia, human immunodeficiency virus and influenza) confirmed, or community-acquired pneumonia with resolved follow-up chest CT findings after treatment.
CT image data acquisition
CT images of the thorax were acquired using the automatic exposure control setting and scan range, and the noise index of was 12.3. CT scans were performed ≤7 days after symptom onset on a helical 64-slice CT GE (Lightspeed Ultra 16, USA; 1.25 mm slice thickness; 1.5 pitch; 120 kVP tube voltage; 100-200 mAs tube current; sagittal and coronal reconstruction thickness, 3 mm with 3-mm intervals) or Siemens (Somatom Definition AS, Germany; 1 mm slice thickness; 1.2 pitch; 120 kVP tube voltage; 100-200 mAs tube current; sagittal and coronal reconstruction thickness, 3 mm with 3-mm intervals; and a sharp reconstruction kernel).
CT image analysis
We summarized several significant COVID-19 CT image features by reviewing recently reported papers published or e-published on chest CT findings from the COVID-19 outbreak in China in Table 2. Referring to other CT image signs in viral pneumonia[16, 17] or community-acquired pneumonia[16, 18], we set seven positive signs from significant COVID-19 image features and four negative signs from significant image features of other non-COVID-19 pneumonia as in Table 3 and Figure 1. In brief, visual scores were defined as follows: score 1, positive significant COVID-19 image features; score -1, non-COVID-19 with viral pneumonia or community-acquired pneumonia image features. An overall score was reached by summing the scores of the eleven features in Table 4.
The image analysis focused on the features of each patient, including (a) number of lobes involved, (b) lesions and distribution characteristics (e.g., peripheral distribution, central distribution, subpleural distribution, and posterior distribution), (c) lesion patterns (e.g., ground glass opacification (GGO) with or without consolidation, crazy-paving pattern, and the shape of the GGO), (d) other signs in the lesion (e.g., bronchial and/or bronchiolar wall thickening), and (e) other findings (e.g., tree-in-bud sign). All CT findings were described according to the Fleischner Society recommendations and similar studies[19-21]. Peripheral distribution was defined as any lesion affecting a peripheral area (3-4 cm in thickness at the lung periphery) with or without central distribution. Central distribution was defined as ONLY central distribution (the central tubular structures in a secondary pulmonary lobule), and any lesion with a peripheral area affected was excluded. Ground glass opacification was defined as hazy opacity that did not obscure the underlying bronchial and vascular margins; consolidation was defined as opacification with obscuration of bronchial structures and pulmonary vessels [19](Fig. 1A, 1B). A crazy-paving pattern is ground-glass opacity superimposed with lines of reticular patterns [22](Fig. 1C). Rounded GGO is a round-shaped GGO in any plane (Fig. 1A). The subpleural bandlike GGO is a pronounced peripheral, subpleural distribution along with axial pleura (Fig. 1D). Central (peribronchovascular) distribution was defined as typically around the bronchiolar vascular bundle and sparing the subpleural surfaces. They are typically at least 5-10 mm away from the pleural surfaces[23]. (Fig. 1E) The tree-in-bud sign was defined as peripheral, small, centrilobular, and well-defined nodules of soft-tissue attenuation connecting to linear, branching opacities that have more than one contiguous branching site[24] (Fig. 1F).
CT images were reviewed retrospectively and independently by two cardiothoracic radiologists (A with 25 years of experience and B with 15 years of experience) who knew that patients had suspected COVID-19 exposure history but were blinded to any other laboratory or RT-PCR data. When a discrepancy of image feature definition and diagnoses existed between the two radiologists, the final result was decided according to their consensus.
Statistical analysis
Continuous variables were presented as medians with interquartile ranges (IQR). Categorical variables were summarized as counts and percentages. Differences between the two groups (confirmed COVID-19 vs. confirmed non-COVID-19) were compared for continuous and categorical variables by a Mann-Whitney U test and chi-squared test, respectively. p<0.05 was considered significant. The receiver operating characteristic (ROC) curve was used to determine the cut-off value of COVID-19 prediction. The area under the curve (AUC) and Youden index were computed. The performance of each cut-off value was evaluated as sensitivity, specificity, positive and negative predictive values. All analyses were performed with MedCalc Statistical Software, version 18.11.3.